view thesis/cortex.org @ 509:81c845a91378

completed initial render at 81 pages, but images are unacceptably scattered.
author Robert McIntyre <rlm@mit.edu>
date Sun, 30 Mar 2014 01:22:23 -0400
parents c11d3fc3e6f0
children f639e2139ce2
line wrap: on
line source
1 #+title: =CORTEX=
2 #+author: Robert McIntyre
3 #+email: rlm@mit.edu
4 #+description: Using embodied AI to facilitate Artificial Imagination.
5 #+keywords: AI, clojure, embodiment
6 #+LaTeX_CLASS_OPTIONS: [nofloat]
8 * COMMENT templates
9 #+caption:
10 #+caption:
11 #+caption:
12 #+caption:
13 #+name: name
14 #+begin_listing clojure
15 #+BEGIN_SRC clojure
16 #+END_SRC
17 #+end_listing
19 #+caption:
20 #+caption:
21 #+caption:
22 #+name: name
23 #+ATTR_LaTeX: :width 10cm
24 [[./images/aurellem-gray.png]]
26 #+caption:
27 #+caption:
28 #+caption:
29 #+caption:
30 #+name: name
31 #+begin_listing clojure
32 #+BEGIN_SRC clojure
33 #+END_SRC
34 #+end_listing
36 #+caption:
37 #+caption:
38 #+caption:
39 #+name: name
40 #+ATTR_LaTeX: :width 10cm
41 [[./images/aurellem-gray.png]]
44 * Empathy and Embodiment as problem solving strategies
46 By the end of this thesis, you will have seen a novel approach to
47 interpreting video using embodiment and empathy. You will have also
48 seen one way to efficiently implement empathy for embodied
49 creatures. Finally, you will become familiar with =CORTEX=, a system
50 for designing and simulating creatures with rich senses, which you
51 may choose to use in your own research.
53 This is the core vision of my thesis: That one of the important ways
54 in which we understand others is by imagining ourselves in their
55 position and emphatically feeling experiences relative to our own
56 bodies. By understanding events in terms of our own previous
57 corporeal experience, we greatly constrain the possibilities of what
58 would otherwise be an unwieldy exponential search. This extra
59 constraint can be the difference between easily understanding what
60 is happening in a video and being completely lost in a sea of
61 incomprehensible color and movement.
63 ** Recognizing actions in video is extremely difficult
65 Consider for example the problem of determining what is happening
66 in a video of which this is one frame:
68 #+caption: A cat drinking some water. Identifying this action is
69 #+caption: beyond the state of the art for computers.
70 #+ATTR_LaTeX: :width 7cm
71 [[./images/cat-drinking.jpg]]
73 It is currently impossible for any computer program to reliably
74 label such a video as ``drinking''. And rightly so -- it is a very
75 hard problem! What features can you describe in terms of low level
76 functions of pixels that can even begin to describe at a high level
77 what is happening here?
79 Or suppose that you are building a program that recognizes chairs.
80 How could you ``see'' the chair in figure \ref{hidden-chair}?
82 #+caption: The chair in this image is quite obvious to humans, but I
83 #+caption: doubt that any modern computer vision program can find it.
84 #+name: hidden-chair
85 #+ATTR_LaTeX: :width 10cm
86 [[./images/fat-person-sitting-at-desk.jpg]]
88 Finally, how is it that you can easily tell the difference between
89 how the girls /muscles/ are working in figure \ref{girl}?
91 #+caption: The mysterious ``common sense'' appears here as you are able
92 #+caption: to discern the difference in how the girl's arm muscles
93 #+caption: are activated between the two images.
94 #+name: girl
95 #+ATTR_LaTeX: :width 7cm
96 [[./images/wall-push.png]]
98 Each of these examples tells us something about what might be going
99 on in our minds as we easily solve these recognition problems.
101 The hidden chairs show us that we are strongly triggered by cues
102 relating to the position of human bodies, and that we can determine
103 the overall physical configuration of a human body even if much of
104 that body is occluded.
106 The picture of the girl pushing against the wall tells us that we
107 have common sense knowledge about the kinetics of our own bodies.
108 We know well how our muscles would have to work to maintain us in
109 most positions, and we can easily project this self-knowledge to
110 imagined positions triggered by images of the human body.
112 ** =EMPATH= neatly solves recognition problems
114 I propose a system that can express the types of recognition
115 problems above in a form amenable to computation. It is split into
116 four parts:
118 - Free/Guided Play :: The creature moves around and experiences the
119 world through its unique perspective. Many otherwise
120 complicated actions are easily described in the language of a
121 full suite of body-centered, rich senses. For example,
122 drinking is the feeling of water sliding down your throat, and
123 cooling your insides. It's often accompanied by bringing your
124 hand close to your face, or bringing your face close to water.
125 Sitting down is the feeling of bending your knees, activating
126 your quadriceps, then feeling a surface with your bottom and
127 relaxing your legs. These body-centered action descriptions
128 can be either learned or hard coded.
129 - Posture Imitation :: When trying to interpret a video or image,
130 the creature takes a model of itself and aligns it with
131 whatever it sees. This alignment can even cross species, as
132 when humans try to align themselves with things like ponies,
133 dogs, or other humans with a different body type.
134 - Empathy :: The alignment triggers associations with
135 sensory data from prior experiences. For example, the
136 alignment itself easily maps to proprioceptive data. Any
137 sounds or obvious skin contact in the video can to a lesser
138 extent trigger previous experience. Segments of previous
139 experiences are stitched together to form a coherent and
140 complete sensory portrait of the scene.
141 - Recognition :: With the scene described in terms of first
142 person sensory events, the creature can now run its
143 action-identification programs on this synthesized sensory
144 data, just as it would if it were actually experiencing the
145 scene first-hand. If previous experience has been accurately
146 retrieved, and if it is analogous enough to the scene, then
147 the creature will correctly identify the action in the scene.
149 For example, I think humans are able to label the cat video as
150 ``drinking'' because they imagine /themselves/ as the cat, and
151 imagine putting their face up against a stream of water and
152 sticking out their tongue. In that imagined world, they can feel
153 the cool water hitting their tongue, and feel the water entering
154 their body, and are able to recognize that /feeling/ as drinking.
155 So, the label of the action is not really in the pixels of the
156 image, but is found clearly in a simulation inspired by those
157 pixels. An imaginative system, having been trained on drinking and
158 non-drinking examples and learning that the most important
159 component of drinking is the feeling of water sliding down one's
160 throat, would analyze a video of a cat drinking in the following
161 manner:
163 1. Create a physical model of the video by putting a ``fuzzy''
164 model of its own body in place of the cat. Possibly also create
165 a simulation of the stream of water.
167 2. Play out this simulated scene and generate imagined sensory
168 experience. This will include relevant muscle contractions, a
169 close up view of the stream from the cat's perspective, and most
170 importantly, the imagined feeling of water entering the
171 mouth. The imagined sensory experience can come from a
172 simulation of the event, but can also be pattern-matched from
173 previous, similar embodied experience.
175 3. The action is now easily identified as drinking by the sense of
176 taste alone. The other senses (such as the tongue moving in and
177 out) help to give plausibility to the simulated action. Note that
178 the sense of vision, while critical in creating the simulation,
179 is not critical for identifying the action from the simulation.
181 For the chair examples, the process is even easier:
183 1. Align a model of your body to the person in the image.
185 2. Generate proprioceptive sensory data from this alignment.
187 3. Use the imagined proprioceptive data as a key to lookup related
188 sensory experience associated with that particular proproceptive
189 feeling.
191 4. Retrieve the feeling of your bottom resting on a surface, your
192 knees bent, and your leg muscles relaxed.
194 5. This sensory information is consistent with the =sitting?=
195 sensory predicate, so you (and the entity in the image) must be
196 sitting.
198 6. There must be a chair-like object since you are sitting.
200 Empathy offers yet another alternative to the age-old AI
201 representation question: ``What is a chair?'' --- A chair is the
202 feeling of sitting.
204 My program, =EMPATH= uses this empathic problem solving technique
205 to interpret the actions of a simple, worm-like creature.
207 #+caption: The worm performs many actions during free play such as
208 #+caption: curling, wiggling, and resting.
209 #+name: worm-intro
210 #+ATTR_LaTeX: :width 15cm
211 [[./images/worm-intro-white.png]]
213 #+caption: =EMPATH= recognized and classified each of these
214 #+caption: poses by inferring the complete sensory experience
215 #+caption: from proprioceptive data.
216 #+name: worm-recognition-intro
217 #+ATTR_LaTeX: :width 15cm
218 [[./images/worm-poses.png]]
220 One powerful advantage of empathic problem solving is that it
221 factors the action recognition problem into two easier problems. To
222 use empathy, you need an /aligner/, which takes the video and a
223 model of your body, and aligns the model with the video. Then, you
224 need a /recognizer/, which uses the aligned model to interpret the
225 action. The power in this method lies in the fact that you describe
226 all actions form a body-centered viewpoint. You are less tied to
227 the particulars of any visual representation of the actions. If you
228 teach the system what ``running'' is, and you have a good enough
229 aligner, the system will from then on be able to recognize running
230 from any point of view, even strange points of view like above or
231 underneath the runner. This is in contrast to action recognition
232 schemes that try to identify actions using a non-embodied approach.
233 If these systems learn about running as viewed from the side, they
234 will not automatically be able to recognize running from any other
235 viewpoint.
237 Another powerful advantage is that using the language of multiple
238 body-centered rich senses to describe body-centerd actions offers a
239 massive boost in descriptive capability. Consider how difficult it
240 would be to compose a set of HOG filters to describe the action of
241 a simple worm-creature ``curling'' so that its head touches its
242 tail, and then behold the simplicity of describing thus action in a
243 language designed for the task (listing \ref{grand-circle-intro}):
245 #+caption: Body-centerd actions are best expressed in a body-centered
246 #+caption: language. This code detects when the worm has curled into a
247 #+caption: full circle. Imagine how you would replicate this functionality
248 #+caption: using low-level pixel features such as HOG filters!
249 #+name: grand-circle-intro
250 #+begin_listing clojure
251 #+begin_src clojure
252 (defn grand-circle?
253 "Does the worm form a majestic circle (one end touching the other)?"
254 [experiences]
255 (and (curled? experiences)
256 (let [worm-touch (:touch (peek experiences))
257 tail-touch (worm-touch 0)
258 head-touch (worm-touch 4)]
259 (and (< 0.2 (contact worm-segment-bottom-tip tail-touch))
260 (< 0.2 (contact worm-segment-top-tip head-touch))))))
261 #+end_src
262 #+end_listing
264 ** =CORTEX= is a toolkit for building sensate creatures
266 I built =CORTEX= to be a general AI research platform for doing
267 experiments involving multiple rich senses and a wide variety and
268 number of creatures. I intend it to be useful as a library for many
269 more projects than just this thesis. =CORTEX= was necessary to meet
270 a need among AI researchers at CSAIL and beyond, which is that
271 people often will invent neat ideas that are best expressed in the
272 language of creatures and senses, but in order to explore those
273 ideas they must first build a platform in which they can create
274 simulated creatures with rich senses! There are many ideas that
275 would be simple to execute (such as =EMPATH=), but attached to them
276 is the multi-month effort to make a good creature simulator. Often,
277 that initial investment of time proves to be too much, and the
278 project must make do with a lesser environment.
280 =CORTEX= is well suited as an environment for embodied AI research
281 for three reasons:
283 - You can create new creatures using Blender, a popular 3D modeling
284 program. Each sense can be specified using special blender nodes
285 with biologically inspired paramaters. You need not write any
286 code to create a creature, and can use a wide library of
287 pre-existing blender models as a base for your own creatures.
289 - =CORTEX= implements a wide variety of senses, including touch,
290 proprioception, vision, hearing, and muscle tension. Complicated
291 senses like touch, and vision involve multiple sensory elements
292 embedded in a 2D surface. You have complete control over the
293 distribution of these sensor elements through the use of simple
294 png image files. In particular, =CORTEX= implements more
295 comprehensive hearing than any other creature simulation system
296 available.
298 - =CORTEX= supports any number of creatures and any number of
299 senses. Time in =CORTEX= dialates so that the simulated creatures
300 always precieve a perfectly smooth flow of time, regardless of
301 the actual computational load.
303 =CORTEX= is built on top of =jMonkeyEngine3=, which is a video game
304 engine designed to create cross-platform 3D desktop games. =CORTEX=
305 is mainly written in clojure, a dialect of =LISP= that runs on the
306 java virtual machine (JVM). The API for creating and simulating
307 creatures and senses is entirely expressed in clojure, though many
308 senses are implemented at the layer of jMonkeyEngine or below. For
309 example, for the sense of hearing I use a layer of clojure code on
310 top of a layer of java JNI bindings that drive a layer of =C++=
311 code which implements a modified version of =OpenAL= to support
312 multiple listeners. =CORTEX= is the only simulation environment
313 that I know of that can support multiple entities that can each
314 hear the world from their own perspective. Other senses also
315 require a small layer of Java code. =CORTEX= also uses =bullet=, a
316 physics simulator written in =C=.
318 #+caption: Here is the worm from above modeled in Blender, a free
319 #+caption: 3D-modeling program. Senses and joints are described
320 #+caption: using special nodes in Blender.
321 #+name: worm-recognition-intro
322 #+ATTR_LaTeX: :width 12cm
323 [[./images/blender-worm.png]]
325 Here are some thing I anticipate that =CORTEX= might be used for:
327 - exploring new ideas about sensory integration
328 - distributed communication among swarm creatures
329 - self-learning using free exploration,
330 - evolutionary algorithms involving creature construction
331 - exploration of exoitic senses and effectors that are not possible
332 in the real world (such as telekenisis or a semantic sense)
333 - imagination using subworlds
335 During one test with =CORTEX=, I created 3,000 creatures each with
336 their own independent senses and ran them all at only 1/80 real
337 time. In another test, I created a detailed model of my own hand,
338 equipped with a realistic distribution of touch (more sensitive at
339 the fingertips), as well as eyes and ears, and it ran at around 1/4
340 real time.
342 #+BEGIN_LaTeX
343 \begin{sidewaysfigure}
344 \includegraphics[width=9.5in]{images/full-hand.png}
345 \caption{
346 I modeled my own right hand in Blender and rigged it with all the
347 senses that {\tt CORTEX} supports. My simulated hand has a
348 biologically inspired distribution of touch sensors. The senses are
349 displayed on the right, and the simulation is displayed on the
350 left. Notice that my hand is curling its fingers, that it can see
351 its own finger from the eye in its palm, and that it can feel its
352 own thumb touching its palm.}
353 \end{sidewaysfigure}
354 #+END_LaTeX
356 ** Contributions
358 - I built =CORTEX=, a comprehensive platform for embodied AI
359 experiments. =CORTEX= supports many features lacking in other
360 systems, such proper simulation of hearing. It is easy to create
361 new =CORTEX= creatures using Blender, a free 3D modeling program.
363 - I built =EMPATH=, which uses =CORTEX= to identify the actions of
364 a worm-like creature using a computational model of empathy.
366 * Building =CORTEX=
368 I intend for =CORTEX= to be used as a general-purpose library for
369 building creatures and outfitting them with senses, so that it will
370 be useful for other researchers who want to test out ideas of their
371 own. To this end, wherver I have had to make archetictural choices
372 about =CORTEX=, I have chosen to give as much freedom to the user as
373 possible, so that =CORTEX= may be used for things I have not
374 forseen.
376 ** Simulation or Reality?
378 The most important archetictural decision of all is the choice to
379 use a computer-simulated environemnt in the first place! The world
380 is a vast and rich place, and for now simulations are a very poor
381 reflection of its complexity. It may be that there is a significant
382 qualatative difference between dealing with senses in the real
383 world and dealing with pale facilimilies of them in a simulation.
384 What are the advantages and disadvantages of a simulation vs.
385 reality?
387 *** Simulation
389 The advantages of virtual reality are that when everything is a
390 simulation, experiments in that simulation are absolutely
391 reproducible. It's also easier to change the character and world
392 to explore new situations and different sensory combinations.
394 If the world is to be simulated on a computer, then not only do
395 you have to worry about whether the character's senses are rich
396 enough to learn from the world, but whether the world itself is
397 rendered with enough detail and realism to give enough working
398 material to the character's senses. To name just a few
399 difficulties facing modern physics simulators: destructibility of
400 the environment, simulation of water/other fluids, large areas,
401 nonrigid bodies, lots of objects, smoke. I don't know of any
402 computer simulation that would allow a character to take a rock
403 and grind it into fine dust, then use that dust to make a clay
404 sculpture, at least not without spending years calculating the
405 interactions of every single small grain of dust. Maybe a
406 simulated world with today's limitations doesn't provide enough
407 richness for real intelligence to evolve.
409 *** Reality
411 The other approach for playing with senses is to hook your
412 software up to real cameras, microphones, robots, etc., and let it
413 loose in the real world. This has the advantage of eliminating
414 concerns about simulating the world at the expense of increasing
415 the complexity of implementing the senses. Instead of just
416 grabbing the current rendered frame for processing, you have to
417 use an actual camera with real lenses and interact with photons to
418 get an image. It is much harder to change the character, which is
419 now partly a physical robot of some sort, since doing so involves
420 changing things around in the real world instead of modifying
421 lines of code. While the real world is very rich and definitely
422 provides enough stimulation for intelligence to develop as
423 evidenced by our own existence, it is also uncontrollable in the
424 sense that a particular situation cannot be recreated perfectly or
425 saved for later use. It is harder to conduct science because it is
426 harder to repeat an experiment. The worst thing about using the
427 real world instead of a simulation is the matter of time. Instead
428 of simulated time you get the constant and unstoppable flow of
429 real time. This severely limits the sorts of software you can use
430 to program the AI because all sense inputs must be handled in real
431 time. Complicated ideas may have to be implemented in hardware or
432 may simply be impossible given the current speed of our
433 processors. Contrast this with a simulation, in which the flow of
434 time in the simulated world can be slowed down to accommodate the
435 limitations of the character's programming. In terms of cost,
436 doing everything in software is far cheaper than building custom
437 real-time hardware. All you need is a laptop and some patience.
439 ** Because of Time, simulation is perferable to reality
441 I envision =CORTEX= being used to support rapid prototyping and
442 iteration of ideas. Even if I could put together a well constructed
443 kit for creating robots, it would still not be enough because of
444 the scourge of real-time processing. Anyone who wants to test their
445 ideas in the real world must always worry about getting their
446 algorithms to run fast enough to process information in real time.
447 The need for real time processing only increases if multiple senses
448 are involved. In the extreme case, even simple algorithms will have
449 to be accelerated by ASIC chips or FPGAs, turning what would
450 otherwise be a few lines of code and a 10x speed penality into a
451 multi-month ordeal. For this reason, =CORTEX= supports
452 /time-dialiation/, which scales back the framerate of the
453 simulation in proportion to the amount of processing each frame.
454 From the perspective of the creatures inside the simulation, time
455 always appears to flow at a constant rate, regardless of how
456 complicated the envorimnent becomes or how many creatures are in
457 the simulation. The cost is that =CORTEX= can sometimes run slower
458 than real time. This can also be an advantage, however ---
459 simulations of very simple creatures in =CORTEX= generally run at
460 40x on my machine!
462 ** What is a sense?
464 If =CORTEX= is to support a wide variety of senses, it would help
465 to have a better understanding of what a ``sense'' actually is!
466 While vision, touch, and hearing all seem like they are quite
467 different things, I was supprised to learn during the course of
468 this thesis that they (and all physical senses) can be expressed as
469 exactly the same mathematical object due to a dimensional argument!
471 Human beings are three-dimensional objects, and the nerves that
472 transmit data from our various sense organs to our brain are
473 essentially one-dimensional. This leaves up to two dimensions in
474 which our sensory information may flow. For example, imagine your
475 skin: it is a two-dimensional surface around a three-dimensional
476 object (your body). It has discrete touch sensors embedded at
477 various points, and the density of these sensors corresponds to the
478 sensitivity of that region of skin. Each touch sensor connects to a
479 nerve, all of which eventually are bundled together as they travel
480 up the spinal cord to the brain. Intersect the spinal nerves with a
481 guillotining plane and you will see all of the sensory data of the
482 skin revealed in a roughly circular two-dimensional image which is
483 the cross section of the spinal cord. Points on this image that are
484 close together in this circle represent touch sensors that are
485 /probably/ close together on the skin, although there is of course
486 some cutting and rearrangement that has to be done to transfer the
487 complicated surface of the skin onto a two dimensional image.
489 Most human senses consist of many discrete sensors of various
490 properties distributed along a surface at various densities. For
491 skin, it is Pacinian corpuscles, Meissner's corpuscles, Merkel's
492 disks, and Ruffini's endings, which detect pressure and vibration
493 of various intensities. For ears, it is the stereocilia distributed
494 along the basilar membrane inside the cochlea; each one is
495 sensitive to a slightly different frequency of sound. For eyes, it
496 is rods and cones distributed along the surface of the retina. In
497 each case, we can describe the sense with a surface and a
498 distribution of sensors along that surface.
500 The neat idea is that every human sense can be effectively
501 described in terms of a surface containing embedded sensors. If the
502 sense had any more dimensions, then there wouldn't be enough room
503 in the spinal chord to transmit the information!
505 Therefore, =CORTEX= must support the ability to create objects and
506 then be able to ``paint'' points along their surfaces to describe
507 each sense.
509 Fortunately this idea is already a well known computer graphics
510 technique called called /UV-mapping/. The three-dimensional surface
511 of a model is cut and smooshed until it fits on a two-dimensional
512 image. You paint whatever you want on that image, and when the
513 three-dimensional shape is rendered in a game the smooshing and
514 cutting is reversed and the image appears on the three-dimensional
515 object.
517 To make a sense, interpret the UV-image as describing the
518 distribution of that senses sensors. To get different types of
519 sensors, you can either use a different color for each type of
520 sensor, or use multiple UV-maps, each labeled with that sensor
521 type. I generally use a white pixel to mean the presence of a
522 sensor and a black pixel to mean the absence of a sensor, and use
523 one UV-map for each sensor-type within a given sense.
525 #+CAPTION: The UV-map for an elongated icososphere. The white
526 #+caption: dots each represent a touch sensor. They are dense
527 #+caption: in the regions that describe the tip of the finger,
528 #+caption: and less dense along the dorsal side of the finger
529 #+caption: opposite the tip.
530 #+name: finger-UV
531 #+ATTR_latex: :width 10cm
532 [[./images/finger-UV.png]]
534 #+caption: Ventral side of the UV-mapped finger. Notice the
535 #+caption: density of touch sensors at the tip.
536 #+name: finger-side-view
537 #+ATTR_LaTeX: :width 10cm
538 [[./images/finger-1.png]]
540 ** Video game engines provide ready-made physics and shading
542 I did not need to write my own physics simulation code or shader to
543 build =CORTEX=. Doing so would lead to a system that is impossible
544 for anyone but myself to use anyway. Instead, I use a video game
545 engine as a base and modify it to accomodate the additional needs
546 of =CORTEX=. Video game engines are an ideal starting point to
547 build =CORTEX=, because they are not far from being creature
548 building systems themselves.
550 First off, general purpose video game engines come with a physics
551 engine and lighting / sound system. The physics system provides
552 tools that can be co-opted to serve as touch, proprioception, and
553 muscles. Since some games support split screen views, a good video
554 game engine will allow you to efficiently create multiple cameras
555 in the simulated world that can be used as eyes. Video game systems
556 offer integrated asset management for things like textures and
557 creatures models, providing an avenue for defining creatures. They
558 also understand UV-mapping, since this technique is used to apply a
559 texture to a model. Finally, because video game engines support a
560 large number of users, as long as =CORTEX= doesn't stray too far
561 from the base system, other researchers can turn to this community
562 for help when doing their research.
564 ** =CORTEX= is based on jMonkeyEngine3
566 While preparing to build =CORTEX= I studied several video game
567 engines to see which would best serve as a base. The top contenders
568 were:
570 - [[http://www.idsoftware.com][Quake II]]/[[http://www.bytonic.de/html/jake2.html][Jake2]] :: The Quake II engine was designed by ID
571 software in 1997. All the source code was released by ID
572 software into the Public Domain several years ago, and as a
573 result it has been ported to many different languages. This
574 engine was famous for its advanced use of realistic shading
575 and had decent and fast physics simulation. The main advantage
576 of the Quake II engine is its simplicity, but I ultimately
577 rejected it because the engine is too tied to the concept of a
578 first-person shooter game. One of the problems I had was that
579 there does not seem to be any easy way to attach multiple
580 cameras to a single character. There are also several physics
581 clipping issues that are corrected in a way that only applies
582 to the main character and do not apply to arbitrary objects.
584 - [[http://source.valvesoftware.com/][Source Engine]] :: The Source Engine evolved from the Quake II
585 and Quake I engines and is used by Valve in the Half-Life
586 series of games. The physics simulation in the Source Engine
587 is quite accurate and probably the best out of all the engines
588 I investigated. There is also an extensive community actively
589 working with the engine. However, applications that use the
590 Source Engine must be written in C++, the code is not open, it
591 only runs on Windows, and the tools that come with the SDK to
592 handle models and textures are complicated and awkward to use.
594 - [[http://jmonkeyengine.com/][jMonkeyEngine3]] :: jMonkeyEngine3 is a new library for creating
595 games in Java. It uses OpenGL to render to the screen and uses
596 screengraphs to avoid drawing things that do not appear on the
597 screen. It has an active community and several games in the
598 pipeline. The engine was not built to serve any particular
599 game but is instead meant to be used for any 3D game.
601 I chose jMonkeyEngine3 because it because it had the most features
602 out of all the free projects I looked at, and because I could then
603 write my code in clojure, an implementation of =LISP= that runs on
604 the JVM.
606 ** =CORTEX= uses Blender to create creature models
608 For the simple worm-like creatures I will use later on in this
609 thesis, I could define a simple API in =CORTEX= that would allow
610 one to create boxes, spheres, etc., and leave that API as the sole
611 way to create creatures. However, for =CORTEX= to truly be useful
612 for other projects, it needs a way to construct complicated
613 creatures. If possible, it would be nice to leverage work that has
614 already been done by the community of 3D modelers, or at least
615 enable people who are talented at moedling but not programming to
616 design =CORTEX= creatures.
618 Therefore, I use Blender, a free 3D modeling program, as the main
619 way to create creatures in =CORTEX=. However, the creatures modeled
620 in Blender must also be simple to simulate in jMonkeyEngine3's game
621 engine, and must also be easy to rig with =CORTEX='s senses. I
622 accomplish this with extensive use of Blender's ``empty nodes.''
624 Empty nodes have no mass, physical presence, or appearance, but
625 they can hold metadata and have names. I use a tree structure of
626 empty nodes to specify senses in the following manner:
628 - Create a single top-level empty node whose name is the name of
629 the sense.
630 - Add empty nodes which each contain meta-data relevant to the
631 sense, including a UV-map describing the number/distribution of
632 sensors if applicable.
633 - Make each empty-node the child of the top-level node.
635 #+caption: An example of annoting a creature model with empty
636 #+caption: nodes to describe the layout of senses. There are
637 #+caption: multiple empty nodes which each describe the position
638 #+caption: of muscles, ears, eyes, or joints.
639 #+name: sense-nodes
640 #+ATTR_LaTeX: :width 10cm
641 [[./images/empty-sense-nodes.png]]
643 ** Bodies are composed of segments connected by joints
645 Blender is a general purpose animation tool, which has been used in
646 the past to create high quality movies such as Sintel
647 \cite{blender}. Though Blender can model and render even complicated
648 things like water, it is crucual to keep models that are meant to
649 be simulated as creatures simple. =Bullet=, which =CORTEX= uses
650 though jMonkeyEngine3, is a rigid-body physics system. This offers
651 a compromise between the expressiveness of a game level and the
652 speed at which it can be simulated, and it means that creatures
653 should be naturally expressed as rigid components held together by
654 joint constraints.
656 But humans are more like a squishy bag with wrapped around some
657 hard bones which define the overall shape. When we move, our skin
658 bends and stretches to accomodate the new positions of our bones.
660 One way to make bodies composed of rigid pieces connected by joints
661 /seem/ more human-like is to use an /armature/, (or /rigging/)
662 system, which defines a overall ``body mesh'' and defines how the
663 mesh deforms as a function of the position of each ``bone'' which
664 is a standard rigid body. This technique is used extensively to
665 model humans and create realistic animations. It is not a good
666 technique for physical simulation, however because it creates a lie
667 -- the skin is not a physical part of the simulation and does not
668 interact with any objects in the world or itself. Objects will pass
669 right though the skin until they come in contact with the
670 underlying bone, which is a physical object. Whithout simulating
671 the skin, the sense of touch has little meaning, and the creature's
672 own vision will lie to it about the true extent of its body.
673 Simulating the skin as a physical object requires some way to
674 continuously update the physical model of the skin along with the
675 movement of the bones, which is unacceptably slow compared to rigid
676 body simulation.
678 Therefore, instead of using the human-like ``deformable bag of
679 bones'' approach, I decided to base my body plans on multiple solid
680 objects that are connected by joints, inspired by the robot =EVE=
681 from the movie WALL-E.
683 #+caption: =EVE= from the movie WALL-E. This body plan turns
684 #+caption: out to be much better suited to my purposes than a more
685 #+caption: human-like one.
686 #+ATTR_LaTeX: :width 10cm
687 [[./images/Eve.jpg]]
689 =EVE='s body is composed of several rigid components that are held
690 together by invisible joint constraints. This is what I mean by
691 ``eve-like''. The main reason that I use eve-style bodies is for
692 efficiency, and so that there will be correspondence between the
693 AI's semses and the physical presence of its body. Each individual
694 section is simulated by a separate rigid body that corresponds
695 exactly with its visual representation and does not change.
696 Sections are connected by invisible joints that are well supported
697 in jMonkeyEngine3. Bullet, the physics backend for jMonkeyEngine3,
698 can efficiently simulate hundreds of rigid bodies connected by
699 joints. Just because sections are rigid does not mean they have to
700 stay as one piece forever; they can be dynamically replaced with
701 multiple sections to simulate splitting in two. This could be used
702 to simulate retractable claws or =EVE='s hands, which are able to
703 coalesce into one object in the movie.
705 *** Solidifying/Connecting a body
707 =CORTEX= creates a creature in two steps: first, it traverses the
708 nodes in the blender file and creates physical representations for
709 any of them that have mass defined in their blender meta-data.
711 #+caption: Program for iterating through the nodes in a blender file
712 #+caption: and generating physical jMonkeyEngine3 objects with mass
713 #+caption: and a matching physics shape.
714 #+name: name
715 #+begin_listing clojure
716 #+begin_src clojure
717 (defn physical!
718 "Iterate through the nodes in creature and make them real physical
719 objects in the simulation."
720 [#^Node creature]
721 (dorun
722 (map
723 (fn [geom]
724 (let [physics-control
725 (RigidBodyControl.
726 (HullCollisionShape.
727 (.getMesh geom))
728 (if-let [mass (meta-data geom "mass")]
729 (float mass) (float 1)))]
730 (.addControl geom physics-control)))
731 (filter #(isa? (class %) Geometry )
732 (node-seq creature)))))
733 #+end_src
734 #+end_listing
736 The next step to making a proper body is to connect those pieces
737 together with joints. jMonkeyEngine has a large array of joints
738 available via =bullet=, such as Point2Point, Cone, Hinge, and a
739 generic Six Degree of Freedom joint, with or without spring
740 restitution.
742 Joints are treated a lot like proper senses, in that there is a
743 top-level empty node named ``joints'' whose children each
744 represent a joint.
746 #+caption: View of the hand model in Blender showing the main ``joints''
747 #+caption: node (highlighted in yellow) and its children which each
748 #+caption: represent a joint in the hand. Each joint node has metadata
749 #+caption: specifying what sort of joint it is.
750 #+name: blender-hand
751 #+ATTR_LaTeX: :width 10cm
752 [[./images/hand-screenshot1.png]]
755 =CORTEX='s procedure for binding the creature together with joints
756 is as follows:
758 - Find the children of the ``joints'' node.
759 - Determine the two spatials the joint is meant to connect.
760 - Create the joint based on the meta-data of the empty node.
762 The higher order function =sense-nodes= from =cortex.sense=
763 simplifies finding the joints based on their parent ``joints''
764 node.
766 #+caption: Retrieving the children empty nodes from a single
767 #+caption: named empty node is a common pattern in =CORTEX=
768 #+caption: further instances of this technique for the senses
769 #+caption: will be omitted
770 #+name: get-empty-nodes
771 #+begin_listing clojure
772 #+begin_src clojure
773 (defn sense-nodes
774 "For some senses there is a special empty blender node whose
775 children are considered markers for an instance of that sense. This
776 function generates functions to find those children, given the name
777 of the special parent node."
778 [parent-name]
779 (fn [#^Node creature]
780 (if-let [sense-node (.getChild creature parent-name)]
781 (seq (.getChildren sense-node)) [])))
783 (def
784 ^{:doc "Return the children of the creature's \"joints\" node."
785 :arglists '([creature])}
786 joints
787 (sense-nodes "joints"))
788 #+end_src
789 #+end_listing
791 To find a joint's targets, =CORTEX= creates a small cube, centered
792 around the empty-node, and grows the cube exponentially until it
793 intersects two physical objects. The objects are ordered according
794 to the joint's rotation, with the first one being the object that
795 has more negative coordinates in the joint's reference frame.
796 Since the objects must be physical, the empty-node itself escapes
797 detection. Because the objects must be physical, =joint-targets=
798 must be called /after/ =physical!= is called.
800 #+caption: Program to find the targets of a joint node by
801 #+caption: exponentiallly growth of a search cube.
802 #+name: joint-targets
803 #+begin_listing clojure
804 #+begin_src clojure
805 (defn joint-targets
806 "Return the two closest two objects to the joint object, ordered
807 from bottom to top according to the joint's rotation."
808 [#^Node parts #^Node joint]
809 (loop [radius (float 0.01)]
810 (let [results (CollisionResults.)]
811 (.collideWith
812 parts
813 (BoundingBox. (.getWorldTranslation joint)
814 radius radius radius) results)
815 (let [targets
816 (distinct
817 (map #(.getGeometry %) results))]
818 (if (>= (count targets) 2)
819 (sort-by
820 #(let [joint-ref-frame-position
821 (jme-to-blender
822 (.mult
823 (.inverse (.getWorldRotation joint))
824 (.subtract (.getWorldTranslation %)
825 (.getWorldTranslation joint))))]
826 (.dot (Vector3f. 1 1 1) joint-ref-frame-position))
827 (take 2 targets))
828 (recur (float (* radius 2))))))))
829 #+end_src
830 #+end_listing
832 Once =CORTEX= finds all joints and targets, it creates them using
833 a dispatch on the metadata of each joint node.
835 #+caption: Program to dispatch on blender metadata and create joints
836 #+caption: sutiable for physical simulation.
837 #+name: joint-dispatch
838 #+begin_listing clojure
839 #+begin_src clojure
840 (defmulti joint-dispatch
841 "Translate blender pseudo-joints into real JME joints."
842 (fn [constraints & _]
843 (:type constraints)))
845 (defmethod joint-dispatch :point
846 [constraints control-a control-b pivot-a pivot-b rotation]
847 (doto (SixDofJoint. control-a control-b pivot-a pivot-b false)
848 (.setLinearLowerLimit Vector3f/ZERO)
849 (.setLinearUpperLimit Vector3f/ZERO)))
851 (defmethod joint-dispatch :hinge
852 [constraints control-a control-b pivot-a pivot-b rotation]
853 (let [axis (if-let [axis (:axis constraints)] axis Vector3f/UNIT_X)
854 [limit-1 limit-2] (:limit constraints)
855 hinge-axis (.mult rotation (blender-to-jme axis))]
856 (doto (HingeJoint. control-a control-b pivot-a pivot-b
857 hinge-axis hinge-axis)
858 (.setLimit limit-1 limit-2))))
860 (defmethod joint-dispatch :cone
861 [constraints control-a control-b pivot-a pivot-b rotation]
862 (let [limit-xz (:limit-xz constraints)
863 limit-xy (:limit-xy constraints)
864 twist (:twist constraints)]
865 (doto (ConeJoint. control-a control-b pivot-a pivot-b
866 rotation rotation)
867 (.setLimit (float limit-xz) (float limit-xy)
868 (float twist)))))
869 #+end_src
870 #+end_listing
872 All that is left for joints it to combine the above pieces into a
873 something that can operate on the collection of nodes that a
874 blender file represents.
876 #+caption: Program to completely create a joint given information
877 #+caption: from a blender file.
878 #+name: connect
879 #+begin_listing clojure
880 #+begin_src clojure
881 (defn connect
882 "Create a joint between 'obj-a and 'obj-b at the location of
883 'joint. The type of joint is determined by the metadata on 'joint.
885 Here are some examples:
886 {:type :point}
887 {:type :hinge :limit [0 (/ Math/PI 2)] :axis (Vector3f. 0 1 0)}
888 (:axis defaults to (Vector3f. 1 0 0) if not provided for hinge joints)
890 {:type :cone :limit-xz 0]
891 :limit-xy 0]
892 :twist 0]} (use XZY rotation mode in blender!)"
893 [#^Node obj-a #^Node obj-b #^Node joint]
894 (let [control-a (.getControl obj-a RigidBodyControl)
895 control-b (.getControl obj-b RigidBodyControl)
896 joint-center (.getWorldTranslation joint)
897 joint-rotation (.toRotationMatrix (.getWorldRotation joint))
898 pivot-a (world-to-local obj-a joint-center)
899 pivot-b (world-to-local obj-b joint-center)]
900 (if-let
901 [constraints (map-vals eval (read-string (meta-data joint "joint")))]
902 ;; A side-effect of creating a joint registers
903 ;; it with both physics objects which in turn
904 ;; will register the joint with the physics system
905 ;; when the simulation is started.
906 (joint-dispatch constraints
907 control-a control-b
908 pivot-a pivot-b
909 joint-rotation))))
910 #+end_src
911 #+end_listing
913 In general, whenever =CORTEX= exposes a sense (or in this case
914 physicality), it provides a function of the type =sense!=, which
915 takes in a collection of nodes and augments it to support that
916 sense. The function returns any controlls necessary to use that
917 sense. In this case =body!= cerates a physical body and returns no
918 control functions.
920 #+caption: Program to give joints to a creature.
921 #+name: name
922 #+begin_listing clojure
923 #+begin_src clojure
924 (defn joints!
925 "Connect the solid parts of the creature with physical joints. The
926 joints are taken from the \"joints\" node in the creature."
927 [#^Node creature]
928 (dorun
929 (map
930 (fn [joint]
931 (let [[obj-a obj-b] (joint-targets creature joint)]
932 (connect obj-a obj-b joint)))
933 (joints creature))))
934 (defn body!
935 "Endow the creature with a physical body connected with joints. The
936 particulars of the joints and the masses of each body part are
937 determined in blender."
938 [#^Node creature]
939 (physical! creature)
940 (joints! creature))
941 #+end_src
942 #+end_listing
944 All of the code you have just seen amounts to only 130 lines, yet
945 because it builds on top of Blender and jMonkeyEngine3, those few
946 lines pack quite a punch!
948 The hand from figure \ref{blender-hand}, which was modeled after
949 my own right hand, can now be given joints and simulated as a
950 creature.
952 #+caption: With the ability to create physical creatures from blender,
953 #+caption: =CORTEX= gets one step closer to becomming a full creature
954 #+caption: simulation environment.
955 #+name: name
956 #+ATTR_LaTeX: :width 15cm
957 [[./images/physical-hand.png]]
959 ** Eyes reuse standard video game components
961 Vision is one of the most important senses for humans, so I need to
962 build a simulated sense of vision for my AI. I will do this with
963 simulated eyes. Each eye can be independently moved and should see
964 its own version of the world depending on where it is.
966 Making these simulated eyes a reality is simple because
967 jMonkeyEngine already contains extensive support for multiple views
968 of the same 3D simulated world. The reason jMonkeyEngine has this
969 support is because the support is necessary to create games with
970 split-screen views. Multiple views are also used to create
971 efficient pseudo-reflections by rendering the scene from a certain
972 perspective and then projecting it back onto a surface in the 3D
973 world.
975 #+caption: jMonkeyEngine supports multiple views to enable
976 #+caption: split-screen games, like GoldenEye, which was one of
977 #+caption: the first games to use split-screen views.
978 #+name: name
979 #+ATTR_LaTeX: :width 10cm
980 [[./images/goldeneye-4-player.png]]
982 *** A Brief Description of jMonkeyEngine's Rendering Pipeline
984 jMonkeyEngine allows you to create a =ViewPort=, which represents a
985 view of the simulated world. You can create as many of these as you
986 want. Every frame, the =RenderManager= iterates through each
987 =ViewPort=, rendering the scene in the GPU. For each =ViewPort= there
988 is a =FrameBuffer= which represents the rendered image in the GPU.
990 #+caption: =ViewPorts= are cameras in the world. During each frame,
991 #+caption: the =RenderManager= records a snapshot of what each view
992 #+caption: is currently seeing; these snapshots are =FrameBuffer= objects.
993 #+name: rendermanagers
994 #+ATTR_LaTeX: :width 10cm
995 [[./images/diagram_rendermanager2.png]]
997 Each =ViewPort= can have any number of attached =SceneProcessor=
998 objects, which are called every time a new frame is rendered. A
999 =SceneProcessor= receives its =ViewPort's= =FrameBuffer= and can do
1000 whatever it wants to the data. Often this consists of invoking GPU
1001 specific operations on the rendered image. The =SceneProcessor= can
1002 also copy the GPU image data to RAM and process it with the CPU.
1004 *** Appropriating Views for Vision
1006 Each eye in the simulated creature needs its own =ViewPort= so
1007 that it can see the world from its own perspective. To this
1008 =ViewPort=, I add a =SceneProcessor= that feeds the visual data to
1009 any arbitrary continuation function for further processing. That
1010 continuation function may perform both CPU and GPU operations on
1011 the data. To make this easy for the continuation function, the
1012 =SceneProcessor= maintains appropriately sized buffers in RAM to
1013 hold the data. It does not do any copying from the GPU to the CPU
1014 itself because it is a slow operation.
1016 #+caption: Function to make the rendered secne in jMonkeyEngine
1017 #+caption: available for further processing.
1018 #+name: pipeline-1
1019 #+begin_listing clojure
1020 #+begin_src clojure
1021 (defn vision-pipeline
1022 "Create a SceneProcessor object which wraps a vision processing
1023 continuation function. The continuation is a function that takes
1024 [#^Renderer r #^FrameBuffer fb #^ByteBuffer b #^BufferedImage bi],
1025 each of which has already been appropriately sized."
1026 [continuation]
1027 (let [byte-buffer (atom nil)
1028 renderer (atom nil)
1029 image (atom nil)]
1030 (proxy [SceneProcessor] []
1031 (initialize
1032 [renderManager viewPort]
1033 (let [cam (.getCamera viewPort)
1034 width (.getWidth cam)
1035 height (.getHeight cam)]
1036 (reset! renderer (.getRenderer renderManager))
1037 (reset! byte-buffer
1038 (BufferUtils/createByteBuffer
1039 (* width height 4)))
1040 (reset! image (BufferedImage.
1041 width height
1042 BufferedImage/TYPE_4BYTE_ABGR))))
1043 (isInitialized [] (not (nil? @byte-buffer)))
1044 (reshape [_ _ _])
1045 (preFrame [_])
1046 (postQueue [_])
1047 (postFrame
1048 [#^FrameBuffer fb]
1049 (.clear @byte-buffer)
1050 (continuation @renderer fb @byte-buffer @image))
1051 (cleanup []))))
1052 #+end_src
1053 #+end_listing
1055 The continuation function given to =vision-pipeline= above will be
1056 given a =Renderer= and three containers for image data. The
1057 =FrameBuffer= references the GPU image data, but the pixel data
1058 can not be used directly on the CPU. The =ByteBuffer= and
1059 =BufferedImage= are initially "empty" but are sized to hold the
1060 data in the =FrameBuffer=. I call transferring the GPU image data
1061 to the CPU structures "mixing" the image data.
1063 *** Optical sensor arrays are described with images and referenced with metadata
1065 The vision pipeline described above handles the flow of rendered
1066 images. Now, =CORTEX= needs simulated eyes to serve as the source
1067 of these images.
1069 An eye is described in blender in the same way as a joint. They
1070 are zero dimensional empty objects with no geometry whose local
1071 coordinate system determines the orientation of the resulting eye.
1072 All eyes are children of a parent node named "eyes" just as all
1073 joints have a parent named "joints". An eye binds to the nearest
1074 physical object with =bind-sense=.
1076 #+caption: Here, the camera is created based on metadata on the
1077 #+caption: eye-node and attached to the nearest physical object
1078 #+caption: with =bind-sense=
1079 #+name: add-eye
1080 #+begin_listing clojure
1081 (defn add-eye!
1082 "Create a Camera centered on the current position of 'eye which
1083 follows the closest physical node in 'creature. The camera will
1084 point in the X direction and use the Z vector as up as determined
1085 by the rotation of these vectors in blender coordinate space. Use
1086 XZY rotation for the node in blender."
1087 [#^Node creature #^Spatial eye]
1088 (let [target (closest-node creature eye)
1089 [cam-width cam-height]
1090 ;;[640 480] ;; graphics card on laptop doesn't support
1091 ;; arbitray dimensions.
1092 (eye-dimensions eye)
1093 cam (Camera. cam-width cam-height)
1094 rot (.getWorldRotation eye)]
1095 (.setLocation cam (.getWorldTranslation eye))
1096 (.lookAtDirection
1097 cam ; this part is not a mistake and
1098 (.mult rot Vector3f/UNIT_X) ; is consistent with using Z in
1099 (.mult rot Vector3f/UNIT_Y)) ; blender as the UP vector.
1100 (.setFrustumPerspective
1101 cam (float 45)
1102 (float (/ (.getWidth cam) (.getHeight cam)))
1103 (float 1)
1104 (float 1000))
1105 (bind-sense target cam) cam))
1106 #+end_listing
1108 *** Simulated Retina
1110 An eye is a surface (the retina) which contains many discrete
1111 sensors to detect light. These sensors can have different
1112 light-sensing properties. In humans, each discrete sensor is
1113 sensitive to red, blue, green, or gray. These different types of
1114 sensors can have different spatial distributions along the retina.
1115 In humans, there is a fovea in the center of the retina which has
1116 a very high density of color sensors, and a blind spot which has
1117 no sensors at all. Sensor density decreases in proportion to
1118 distance from the fovea.
1120 I want to be able to model any retinal configuration, so my
1121 eye-nodes in blender contain metadata pointing to images that
1122 describe the precise position of the individual sensors using
1123 white pixels. The meta-data also describes the precise sensitivity
1124 to light that the sensors described in the image have. An eye can
1125 contain any number of these images. For example, the metadata for
1126 an eye might look like this:
1128 #+begin_src clojure
1129 {0xFF0000 "Models/test-creature/retina-small.png"}
1130 #+end_src
1132 #+caption: An example retinal profile image. White pixels are
1133 #+caption: photo-sensitive elements. The distribution of white
1134 #+caption: pixels is denser in the middle and falls off at the
1135 #+caption: edges and is inspired by the human retina.
1136 #+name: retina
1137 #+ATTR_LaTeX: :width 10cm
1138 [[./images/retina-small.png]]
1140 Together, the number 0xFF0000 and the image image above describe
1141 the placement of red-sensitive sensory elements.
1143 Meta-data to very crudely approximate a human eye might be
1144 something like this:
1146 #+begin_src clojure
1147 (let [retinal-profile "Models/test-creature/retina-small.png"]
1148 {0xFF0000 retinal-profile
1149 0x00FF00 retinal-profile
1150 0x0000FF retinal-profile
1151 0xFFFFFF retinal-profile})
1152 #+end_src
1154 The numbers that serve as keys in the map determine a sensor's
1155 relative sensitivity to the channels red, green, and blue. These
1156 sensitivity values are packed into an integer in the order
1157 =|_|R|G|B|= in 8-bit fields. The RGB values of a pixel in the
1158 image are added together with these sensitivities as linear
1159 weights. Therefore, 0xFF0000 means sensitive to red only while
1160 0xFFFFFF means sensitive to all colors equally (gray).
1162 #+caption: This is the core of vision in =CORTEX=. A given eye node
1163 #+caption: is converted into a function that returns visual
1164 #+caption: information from the simulation.
1165 #+name: vision-kernel
1166 #+begin_listing clojure
1167 #+BEGIN_SRC clojure
1168 (defn vision-kernel
1169 "Returns a list of functions, each of which will return a color
1170 channel's worth of visual information when called inside a running
1171 simulation."
1172 [#^Node creature #^Spatial eye & {skip :skip :or {skip 0}}]
1173 (let [retinal-map (retina-sensor-profile eye)
1174 camera (add-eye! creature eye)
1175 vision-image
1176 (atom
1177 (BufferedImage. (.getWidth camera)
1178 (.getHeight camera)
1179 BufferedImage/TYPE_BYTE_BINARY))
1180 register-eye!
1181 (runonce
1182 (fn [world]
1183 (add-camera!
1184 world camera
1185 (let [counter (atom 0)]
1186 (fn [r fb bb bi]
1187 (if (zero? (rem (swap! counter inc) (inc skip)))
1188 (reset! vision-image
1189 (BufferedImage! r fb bb bi))))))))]
1190 (vec
1191 (map
1192 (fn [[key image]]
1193 (let [whites (white-coordinates image)
1194 topology (vec (collapse whites))
1195 sensitivity (sensitivity-presets key key)]
1196 (attached-viewport.
1197 (fn [world]
1198 (register-eye! world)
1199 (vector
1200 topology
1201 (vec
1202 (for [[x y] whites]
1203 (pixel-sense
1204 sensitivity
1205 (.getRGB @vision-image x y))))))
1206 register-eye!)))
1207 retinal-map))))
1208 #+END_SRC
1209 #+end_listing
1211 Note that since each of the functions generated by =vision-kernel=
1212 shares the same =register-eye!= function, the eye will be
1213 registered only once the first time any of the functions from the
1214 list returned by =vision-kernel= is called. Each of the functions
1215 returned by =vision-kernel= also allows access to the =Viewport=
1216 through which it receives images.
1218 All the hard work has been done; all that remains is to apply
1219 =vision-kernel= to each eye in the creature and gather the results
1220 into one list of functions.
1223 #+caption: With =vision!=, =CORTEX= is already a fine simulation
1224 #+caption: environment for experimenting with different types of
1225 #+caption: eyes.
1226 #+name: vision!
1227 #+begin_listing clojure
1228 #+BEGIN_SRC clojure
1229 (defn vision!
1230 "Returns a list of functions, each of which returns visual sensory
1231 data when called inside a running simulation."
1232 [#^Node creature & {skip :skip :or {skip 0}}]
1233 (reduce
1234 concat
1235 (for [eye (eyes creature)]
1236 (vision-kernel creature eye))))
1237 #+END_SRC
1238 #+end_listing
1240 #+caption: Simulated vision with a test creature and the
1241 #+caption: human-like eye approximation. Notice how each channel
1242 #+caption: of the eye responds differently to the differently
1243 #+caption: colored balls.
1244 #+name: worm-vision-test.
1245 #+ATTR_LaTeX: :width 13cm
1246 [[./images/worm-vision.png]]
1248 The vision code is not much more complicated than the body code,
1249 and enables multiple further paths for simulated vision. For
1250 example, it is quite easy to create bifocal vision -- you just
1251 make two eyes next to each other in blender! It is also possible
1252 to encode vision transforms in the retinal files. For example, the
1253 human like retina file in figure \ref{retina} approximates a
1254 log-polar transform.
1256 This vision code has already been absorbed by the jMonkeyEngine
1257 community and is now (in modified form) part of a system for
1258 capturing in-game video to a file.
1260 ** Hearing is hard; =CORTEX= does it right
1262 At the end of this section I will have simulated ears that work the
1263 same way as the simulated eyes in the last section. I will be able to
1264 place any number of ear-nodes in a blender file, and they will bind to
1265 the closest physical object and follow it as it moves around. Each ear
1266 will provide access to the sound data it picks up between every frame.
1268 Hearing is one of the more difficult senses to simulate, because there
1269 is less support for obtaining the actual sound data that is processed
1270 by jMonkeyEngine3. There is no "split-screen" support for rendering
1271 sound from different points of view, and there is no way to directly
1272 access the rendered sound data.
1274 =CORTEX='s hearing is unique because it does not have any
1275 limitations compared to other simulation environments. As far as I
1276 know, there is no other system that supports multiple listerers,
1277 and the sound demo at the end of this section is the first time
1278 it's been done in a video game environment.
1280 *** Brief Description of jMonkeyEngine's Sound System
1282 jMonkeyEngine's sound system works as follows:
1284 - jMonkeyEngine uses the =AppSettings= for the particular
1285 application to determine what sort of =AudioRenderer= should be
1286 used.
1287 - Although some support is provided for multiple AudioRendering
1288 backends, jMonkeyEngine at the time of this writing will either
1289 pick no =AudioRenderer= at all, or the =LwjglAudioRenderer=.
1290 - jMonkeyEngine tries to figure out what sort of system you're
1291 running and extracts the appropriate native libraries.
1292 - The =LwjglAudioRenderer= uses the [[http://lwjgl.org/][=LWJGL=]] (LightWeight Java Game
1293 Library) bindings to interface with a C library called [[http://kcat.strangesoft.net/openal.html][=OpenAL=]]
1294 - =OpenAL= renders the 3D sound and feeds the rendered sound
1295 directly to any of various sound output devices with which it
1296 knows how to communicate.
1298 A consequence of this is that there's no way to access the actual
1299 sound data produced by =OpenAL=. Even worse, =OpenAL= only supports
1300 one /listener/ (it renders sound data from only one perspective),
1301 which normally isn't a problem for games, but becomes a problem
1302 when trying to make multiple AI creatures that can each hear the
1303 world from a different perspective.
1305 To make many AI creatures in jMonkeyEngine that can each hear the
1306 world from their own perspective, or to make a single creature with
1307 many ears, it is necessary to go all the way back to =OpenAL= and
1308 implement support for simulated hearing there.
1310 *** Extending =OpenAl=
1312 Extending =OpenAL= to support multiple listeners requires 500
1313 lines of =C= code and is too hairy to mention here. Instead, I
1314 will show a small amount of extension code and go over the high
1315 level stragety. Full source is of course available with the
1316 =CORTEX= distribution if you're interested.
1318 =OpenAL= goes to great lengths to support many different systems,
1319 all with different sound capabilities and interfaces. It
1320 accomplishes this difficult task by providing code for many
1321 different sound backends in pseudo-objects called /Devices/.
1322 There's a device for the Linux Open Sound System and the Advanced
1323 Linux Sound Architecture, there's one for Direct Sound on Windows,
1324 and there's even one for Solaris. =OpenAL= solves the problem of
1325 platform independence by providing all these Devices.
1327 Wrapper libraries such as LWJGL are free to examine the system on
1328 which they are running and then select an appropriate device for
1329 that system.
1331 There are also a few "special" devices that don't interface with
1332 any particular system. These include the Null Device, which
1333 doesn't do anything, and the Wave Device, which writes whatever
1334 sound it receives to a file, if everything has been set up
1335 correctly when configuring =OpenAL=.
1337 Actual mixing (doppler shift and distance.environment-based
1338 attenuation) of the sound data happens in the Devices, and they
1339 are the only point in the sound rendering process where this data
1340 is available.
1342 Therefore, in order to support multiple listeners, and get the
1343 sound data in a form that the AIs can use, it is necessary to
1344 create a new Device which supports this feature.
1346 Adding a device to OpenAL is rather tricky -- there are five
1347 separate files in the =OpenAL= source tree that must be modified
1348 to do so. I named my device the "Multiple Audio Send" Device, or
1349 =Send= Device for short, since it sends audio data back to the
1350 calling application like an Aux-Send cable on a mixing board.
1352 The main idea behind the Send device is to take advantage of the
1353 fact that LWJGL only manages one /context/ when using OpenAL. A
1354 /context/ is like a container that holds samples and keeps track
1355 of where the listener is. In order to support multiple listeners,
1356 the Send device identifies the LWJGL context as the master
1357 context, and creates any number of slave contexts to represent
1358 additional listeners. Every time the device renders sound, it
1359 synchronizes every source from the master LWJGL context to the
1360 slave contexts. Then, it renders each context separately, using a
1361 different listener for each one. The rendered sound is made
1362 available via JNI to jMonkeyEngine.
1364 Switching between contexts is not the normal operation of a
1365 Device, and one of the problems with doing so is that a Device
1366 normally keeps around a few pieces of state such as the
1367 =ClickRemoval= array above which will become corrupted if the
1368 contexts are not rendered in parallel. The solution is to create a
1369 copy of this normally global device state for each context, and
1370 copy it back and forth into and out of the actual device state
1371 whenever a context is rendered.
1373 The core of the =Send= device is the =syncSources= function, which
1374 does the job of copying all relevant data from one context to
1375 another.
1377 #+caption: Program for extending =OpenAL= to support multiple
1378 #+caption: listeners via context copying/switching.
1379 #+name: sync-openal-sources
1380 #+begin_listing c
1381 #+BEGIN_SRC c
1382 void syncSources(ALsource *masterSource, ALsource *slaveSource,
1383 ALCcontext *masterCtx, ALCcontext *slaveCtx){
1384 ALuint master = masterSource->source;
1385 ALuint slave = slaveSource->source;
1386 ALCcontext *current = alcGetCurrentContext();
1388 syncSourcef(master,slave,masterCtx,slaveCtx,AL_PITCH);
1389 syncSourcef(master,slave,masterCtx,slaveCtx,AL_GAIN);
1390 syncSourcef(master,slave,masterCtx,slaveCtx,AL_MAX_DISTANCE);
1391 syncSourcef(master,slave,masterCtx,slaveCtx,AL_ROLLOFF_FACTOR);
1392 syncSourcef(master,slave,masterCtx,slaveCtx,AL_REFERENCE_DISTANCE);
1393 syncSourcef(master,slave,masterCtx,slaveCtx,AL_MIN_GAIN);
1394 syncSourcef(master,slave,masterCtx,slaveCtx,AL_MAX_GAIN);
1395 syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_OUTER_GAIN);
1396 syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_INNER_ANGLE);
1397 syncSourcef(master,slave,masterCtx,slaveCtx,AL_CONE_OUTER_ANGLE);
1398 syncSourcef(master,slave,masterCtx,slaveCtx,AL_SEC_OFFSET);
1399 syncSourcef(master,slave,masterCtx,slaveCtx,AL_SAMPLE_OFFSET);
1400 syncSourcef(master,slave,masterCtx,slaveCtx,AL_BYTE_OFFSET);
1402 syncSource3f(master,slave,masterCtx,slaveCtx,AL_POSITION);
1403 syncSource3f(master,slave,masterCtx,slaveCtx,AL_VELOCITY);
1404 syncSource3f(master,slave,masterCtx,slaveCtx,AL_DIRECTION);
1406 syncSourcei(master,slave,masterCtx,slaveCtx,AL_SOURCE_RELATIVE);
1407 syncSourcei(master,slave,masterCtx,slaveCtx,AL_LOOPING);
1409 alcMakeContextCurrent(masterCtx);
1410 ALint source_type;
1411 alGetSourcei(master, AL_SOURCE_TYPE, &source_type);
1413 // Only static sources are currently synchronized!
1414 if (AL_STATIC == source_type){
1415 ALint master_buffer;
1416 ALint slave_buffer;
1417 alGetSourcei(master, AL_BUFFER, &master_buffer);
1418 alcMakeContextCurrent(slaveCtx);
1419 alGetSourcei(slave, AL_BUFFER, &slave_buffer);
1420 if (master_buffer != slave_buffer){
1421 alSourcei(slave, AL_BUFFER, master_buffer);
1425 // Synchronize the state of the two sources.
1426 alcMakeContextCurrent(masterCtx);
1427 ALint masterState;
1428 ALint slaveState;
1430 alGetSourcei(master, AL_SOURCE_STATE, &masterState);
1431 alcMakeContextCurrent(slaveCtx);
1432 alGetSourcei(slave, AL_SOURCE_STATE, &slaveState);
1434 if (masterState != slaveState){
1435 switch (masterState){
1436 case AL_INITIAL : alSourceRewind(slave); break;
1437 case AL_PLAYING : alSourcePlay(slave); break;
1438 case AL_PAUSED : alSourcePause(slave); break;
1439 case AL_STOPPED : alSourceStop(slave); break;
1442 // Restore whatever context was previously active.
1443 alcMakeContextCurrent(current);
1445 #+END_SRC
1446 #+end_listing
1448 With this special context-switching device, and some ugly JNI
1449 bindings that are not worth mentioning, =CORTEX= gains the ability
1450 to access multiple sound streams from =OpenAL=.
1452 #+caption: Program to create an ear from a blender empty node. The ear
1453 #+caption: follows around the nearest physical object and passes
1454 #+caption: all sensory data to a continuation function.
1455 #+name: add-ear
1456 #+begin_listing clojure
1457 #+BEGIN_SRC clojure
1458 (defn add-ear!
1459 "Create a Listener centered on the current position of 'ear
1460 which follows the closest physical node in 'creature and
1461 sends sound data to 'continuation."
1462 [#^Application world #^Node creature #^Spatial ear continuation]
1463 (let [target (closest-node creature ear)
1464 lis (Listener.)
1465 audio-renderer (.getAudioRenderer world)
1466 sp (hearing-pipeline continuation)]
1467 (.setLocation lis (.getWorldTranslation ear))
1468 (.setRotation lis (.getWorldRotation ear))
1469 (bind-sense target lis)
1470 (update-listener-velocity! target lis)
1471 (.addListener audio-renderer lis)
1472 (.registerSoundProcessor audio-renderer lis sp)))
1473 #+END_SRC
1474 #+end_listing
1476 The =Send= device, unlike most of the other devices in =OpenAL=,
1477 does not render sound unless asked. This enables the system to
1478 slow down or speed up depending on the needs of the AIs who are
1479 using it to listen. If the device tried to render samples in
1480 real-time, a complicated AI whose mind takes 100 seconds of
1481 computer time to simulate 1 second of AI-time would miss almost
1482 all of the sound in its environment!
1484 #+caption: Program to enable arbitrary hearing in =CORTEX=
1485 #+name: hearing
1486 #+begin_listing clojure
1487 #+BEGIN_SRC clojure
1488 (defn hearing-kernel
1489 "Returns a function which returns auditory sensory data when called
1490 inside a running simulation."
1491 [#^Node creature #^Spatial ear]
1492 (let [hearing-data (atom [])
1493 register-listener!
1494 (runonce
1495 (fn [#^Application world]
1496 (add-ear!
1497 world creature ear
1498 (comp #(reset! hearing-data %)
1499 byteBuffer->pulse-vector))))]
1500 (fn [#^Application world]
1501 (register-listener! world)
1502 (let [data @hearing-data
1503 topology
1504 (vec (map #(vector % 0) (range 0 (count data))))]
1505 [topology data]))))
1507 (defn hearing!
1508 "Endow the creature in a particular world with the sense of
1509 hearing. Will return a sequence of functions, one for each ear,
1510 which when called will return the auditory data from that ear."
1511 [#^Node creature]
1512 (for [ear (ears creature)]
1513 (hearing-kernel creature ear)))
1514 #+END_SRC
1515 #+end_listing
1517 Armed with these functions, =CORTEX= is able to test possibly the
1518 first ever instance of multiple listeners in a video game engine
1519 based simulation!
1521 #+caption: Here a simple creature responds to sound by changing
1522 #+caption: its color from gray to green when the total volume
1523 #+caption: goes over a threshold.
1524 #+name: sound-test
1525 #+begin_listing java
1526 #+BEGIN_SRC java
1527 /**
1528 * Respond to sound! This is the brain of an AI entity that
1529 * hears its surroundings and reacts to them.
1530 */
1531 public void process(ByteBuffer audioSamples,
1532 int numSamples, AudioFormat format) {
1533 audioSamples.clear();
1534 byte[] data = new byte[numSamples];
1535 float[] out = new float[numSamples];
1536 audioSamples.get(data);
1537 FloatSampleTools.
1538 byte2floatInterleaved
1539 (data, 0, out, 0, numSamples/format.getFrameSize(), format);
1541 float max = Float.NEGATIVE_INFINITY;
1542 for (float f : out){if (f > max) max = f;}
1543 audioSamples.clear();
1545 if (max > 0.1){
1546 entity.getMaterial().setColor("Color", ColorRGBA.Green);
1548 else {
1549 entity.getMaterial().setColor("Color", ColorRGBA.Gray);
1551 #+END_SRC
1552 #+end_listing
1554 #+caption: First ever simulation of multiple listerners in =CORTEX=.
1555 #+caption: Each cube is a creature which processes sound data with
1556 #+caption: the =process= function from listing \ref{sound-test}.
1557 #+caption: the ball is constantally emiting a pure tone of
1558 #+caption: constant volume. As it approaches the cubes, they each
1559 #+caption: change color in response to the sound.
1560 #+name: sound-cubes.
1561 #+ATTR_LaTeX: :width 10cm
1562 [[./images/java-hearing-test.png]]
1564 This system of hearing has also been co-opted by the
1565 jMonkeyEngine3 community and is used to record audio for demo
1566 videos.
1568 ** Touch uses hundreds of hair-like elements
1570 Touch is critical to navigation and spatial reasoning and as such I
1571 need a simulated version of it to give to my AI creatures.
1573 Human skin has a wide array of touch sensors, each of which
1574 specialize in detecting different vibrational modes and pressures.
1575 These sensors can integrate a vast expanse of skin (i.e. your
1576 entire palm), or a tiny patch of skin at the tip of your finger.
1577 The hairs of the skin help detect objects before they even come
1578 into contact with the skin proper.
1580 However, touch in my simulated world can not exactly correspond to
1581 human touch because my creatures are made out of completely rigid
1582 segments that don't deform like human skin.
1584 Instead of measuring deformation or vibration, I surround each
1585 rigid part with a plenitude of hair-like objects (/feelers/) which
1586 do not interact with the physical world. Physical objects can pass
1587 through them with no effect. The feelers are able to tell when
1588 other objects pass through them, and they constantly report how
1589 much of their extent is covered. So even though the creature's body
1590 parts do not deform, the feelers create a margin around those body
1591 parts which achieves a sense of touch which is a hybrid between a
1592 human's sense of deformation and sense from hairs.
1594 Implementing touch in jMonkeyEngine follows a different technical
1595 route than vision and hearing. Those two senses piggybacked off
1596 jMonkeyEngine's 3D audio and video rendering subsystems. To
1597 simulate touch, I use jMonkeyEngine's physics system to execute
1598 many small collision detections, one for each feeler. The placement
1599 of the feelers is determined by a UV-mapped image which shows where
1600 each feeler should be on the 3D surface of the body.
1602 *** Defining Touch Meta-Data in Blender
1604 Each geometry can have a single UV map which describes the
1605 position of the feelers which will constitute its sense of touch.
1606 This image path is stored under the ``touch'' key. The image itself
1607 is black and white, with black meaning a feeler length of 0 (no
1608 feeler is present) and white meaning a feeler length of =scale=,
1609 which is a float stored under the key "scale".
1611 #+caption: Touch does not use empty nodes, to store metadata,
1612 #+caption: because the metadata of each solid part of a
1613 #+caption: creature's body is sufficient.
1614 #+name: touch-meta-data
1615 #+begin_listing clojure
1616 #+BEGIN_SRC clojure
1617 (defn tactile-sensor-profile
1618 "Return the touch-sensor distribution image in BufferedImage format,
1619 or nil if it does not exist."
1620 [#^Geometry obj]
1621 (if-let [image-path (meta-data obj "touch")]
1622 (load-image image-path)))
1624 (defn tactile-scale
1625 "Return the length of each feeler. Default scale is 0.01
1626 jMonkeyEngine units."
1627 [#^Geometry obj]
1628 (if-let [scale (meta-data obj "scale")]
1629 scale 0.1))
1630 #+END_SRC
1631 #+end_listing
1633 Here is an example of a UV-map which specifies the position of
1634 touch sensors along the surface of the upper segment of a fingertip.
1636 #+caption: This is the tactile-sensor-profile for the upper segment
1637 #+caption: of a fingertip. It defines regions of high touch sensitivity
1638 #+caption: (where there are many white pixels) and regions of low
1639 #+caption: sensitivity (where white pixels are sparse).
1640 #+name: fingertip-UV
1641 #+ATTR_LaTeX: :width 13cm
1642 [[./images/finger-UV.png]]
1644 *** Implementation Summary
1646 To simulate touch there are three conceptual steps. For each solid
1647 object in the creature, you first have to get UV image and scale
1648 parameter which define the position and length of the feelers.
1649 Then, you use the triangles which comprise the mesh and the UV
1650 data stored in the mesh to determine the world-space position and
1651 orientation of each feeler. Then once every frame, update these
1652 positions and orientations to match the current position and
1653 orientation of the object, and use physics collision detection to
1654 gather tactile data.
1656 Extracting the meta-data has already been described. The third
1657 step, physics collision detection, is handled in =touch-kernel=.
1658 Translating the positions and orientations of the feelers from the
1659 UV-map to world-space is itself a three-step process.
1661 - Find the triangles which make up the mesh in pixel-space and in
1662 world-space. \\(=triangles=, =pixel-triangles=).
1664 - Find the coordinates of each feeler in world-space. These are
1665 the origins of the feelers. (=feeler-origins=).
1667 - Calculate the normals of the triangles in world space, and add
1668 them to each of the origins of the feelers. These are the
1669 normalized coordinates of the tips of the feelers.
1670 (=feeler-tips=).
1672 *** Triangle Math
1674 The rigid objects which make up a creature have an underlying
1675 =Geometry=, which is a =Mesh= plus a =Material= and other
1676 important data involved with displaying the object.
1678 A =Mesh= is composed of =Triangles=, and each =Triangle= has three
1679 vertices which have coordinates in world space and UV space.
1681 Here, =triangles= gets all the world-space triangles which
1682 comprise a mesh, while =pixel-triangles= gets those same triangles
1683 expressed in pixel coordinates (which are UV coordinates scaled to
1684 fit the height and width of the UV image).
1686 #+caption: Programs to extract triangles from a geometry and get
1687 #+caption: their verticies in both world and UV-coordinates.
1688 #+name: get-triangles
1689 #+begin_listing clojure
1690 #+BEGIN_SRC clojure
1691 (defn triangle
1692 "Get the triangle specified by triangle-index from the mesh."
1693 [#^Geometry geo triangle-index]
1694 (triangle-seq
1695 (let [scratch (Triangle.)]
1696 (.getTriangle (.getMesh geo) triangle-index scratch) scratch)))
1698 (defn triangles
1699 "Return a sequence of all the Triangles which comprise a given
1700 Geometry."
1701 [#^Geometry geo]
1702 (map (partial triangle geo) (range (.getTriangleCount (.getMesh geo)))))
1704 (defn triangle-vertex-indices
1705 "Get the triangle vertex indices of a given triangle from a given
1706 mesh."
1707 [#^Mesh mesh triangle-index]
1708 (let [indices (int-array 3)]
1709 (.getTriangle mesh triangle-index indices)
1710 (vec indices)))
1712 (defn vertex-UV-coord
1713 "Get the UV-coordinates of the vertex named by vertex-index"
1714 [#^Mesh mesh vertex-index]
1715 (let [UV-buffer
1716 (.getData
1717 (.getBuffer
1718 mesh
1719 VertexBuffer$Type/TexCoord))]
1720 [(.get UV-buffer (* vertex-index 2))
1721 (.get UV-buffer (+ 1 (* vertex-index 2)))]))
1723 (defn pixel-triangle [#^Geometry geo image index]
1724 (let [mesh (.getMesh geo)
1725 width (.getWidth image)
1726 height (.getHeight image)]
1727 (vec (map (fn [[u v]] (vector (* width u) (* height v)))
1728 (map (partial vertex-UV-coord mesh)
1729 (triangle-vertex-indices mesh index))))))
1731 (defn pixel-triangles
1732 "The pixel-space triangles of the Geometry, in the same order as
1733 (triangles geo)"
1734 [#^Geometry geo image]
1735 (let [height (.getHeight image)
1736 width (.getWidth image)]
1737 (map (partial pixel-triangle geo image)
1738 (range (.getTriangleCount (.getMesh geo))))))
1739 #+END_SRC
1740 #+end_listing
1742 *** The Affine Transform from one Triangle to Another
1744 =pixel-triangles= gives us the mesh triangles expressed in pixel
1745 coordinates and =triangles= gives us the mesh triangles expressed
1746 in world coordinates. The tactile-sensor-profile gives the
1747 position of each feeler in pixel-space. In order to convert
1748 pixel-space coordinates into world-space coordinates we need
1749 something that takes coordinates on the surface of one triangle
1750 and gives the corresponding coordinates on the surface of another
1751 triangle.
1753 Triangles are [[http://mathworld.wolfram.com/AffineTransformation.html ][affine]], which means any triangle can be transformed
1754 into any other by a combination of translation, scaling, and
1755 rotation. The affine transformation from one triangle to another
1756 is readily computable if the triangle is expressed in terms of a
1757 $4x4$ matrix.
1759 #+BEGIN_LaTeX
1760 $$
1761 \begin{bmatrix}
1762 x_1 & x_2 & x_3 & n_x \\
1763 y_1 & y_2 & y_3 & n_y \\
1764 z_1 & z_2 & z_3 & n_z \\
1765 1 & 1 & 1 & 1
1766 \end{bmatrix}
1767 $$
1768 #+END_LaTeX
1770 Here, the first three columns of the matrix are the vertices of
1771 the triangle. The last column is the right-handed unit normal of
1772 the triangle.
1774 With two triangles $T_{1}$ and $T_{2}$ each expressed as a
1775 matrix like above, the affine transform from $T_{1}$ to $T_{2}$
1776 is $T_{2}T_{1}^{-1}$.
1778 The clojure code below recapitulates the formulas above, using
1779 jMonkeyEngine's =Matrix4f= objects, which can describe any affine
1780 transformation.
1782 #+caption: Program to interpert triangles as affine transforms.
1783 #+name: triangle-affine
1784 #+begin_listing clojure
1785 #+BEGIN_SRC clojure
1786 (defn triangle->matrix4f
1787 "Converts the triangle into a 4x4 matrix: The first three columns
1788 contain the vertices of the triangle; the last contains the unit
1789 normal of the triangle. The bottom row is filled with 1s."
1790 [#^Triangle t]
1791 (let [mat (Matrix4f.)
1792 [vert-1 vert-2 vert-3]
1793 (mapv #(.get t %) (range 3))
1794 unit-normal (do (.calculateNormal t)(.getNormal t))
1795 vertices [vert-1 vert-2 vert-3 unit-normal]]
1796 (dorun
1797 (for [row (range 4) col (range 3)]
1798 (do
1799 (.set mat col row (.get (vertices row) col))
1800 (.set mat 3 row 1)))) mat))
1802 (defn triangles->affine-transform
1803 "Returns the affine transformation that converts each vertex in the
1804 first triangle into the corresponding vertex in the second
1805 triangle."
1806 [#^Triangle tri-1 #^Triangle tri-2]
1807 (.mult
1808 (triangle->matrix4f tri-2)
1809 (.invert (triangle->matrix4f tri-1))))
1810 #+END_SRC
1811 #+end_listing
1813 *** Triangle Boundaries
1815 For efficiency's sake I will divide the tactile-profile image into
1816 small squares which inscribe each pixel-triangle, then extract the
1817 points which lie inside the triangle and map them to 3D-space using
1818 =triangle-transform= above. To do this I need a function,
1819 =convex-bounds= which finds the smallest box which inscribes a 2D
1820 triangle.
1822 =inside-triangle?= determines whether a point is inside a triangle
1823 in 2D pixel-space.
1825 #+caption: Program to efficiently determine point includion
1826 #+caption: in a triangle.
1827 #+name: in-triangle
1828 #+begin_listing clojure
1829 #+BEGIN_SRC clojure
1830 (defn convex-bounds
1831 "Returns the smallest square containing the given vertices, as a
1832 vector of integers [left top width height]."
1833 [verts]
1834 (let [xs (map first verts)
1835 ys (map second verts)
1836 x0 (Math/floor (apply min xs))
1837 y0 (Math/floor (apply min ys))
1838 x1 (Math/ceil (apply max xs))
1839 y1 (Math/ceil (apply max ys))]
1840 [x0 y0 (- x1 x0) (- y1 y0)]))
1842 (defn same-side?
1843 "Given the points p1 and p2 and the reference point ref, is point p
1844 on the same side of the line that goes through p1 and p2 as ref is?"
1845 [p1 p2 ref p]
1846 (<=
1848 (.dot
1849 (.cross (.subtract p2 p1) (.subtract p p1))
1850 (.cross (.subtract p2 p1) (.subtract ref p1)))))
1852 (defn inside-triangle?
1853 "Is the point inside the triangle?"
1854 {:author "Dylan Holmes"}
1855 [#^Triangle tri #^Vector3f p]
1856 (let [[vert-1 vert-2 vert-3] [(.get1 tri) (.get2 tri) (.get3 tri)]]
1857 (and
1858 (same-side? vert-1 vert-2 vert-3 p)
1859 (same-side? vert-2 vert-3 vert-1 p)
1860 (same-side? vert-3 vert-1 vert-2 p))))
1861 #+END_SRC
1862 #+end_listing
1864 *** Feeler Coordinates
1866 The triangle-related functions above make short work of
1867 calculating the positions and orientations of each feeler in
1868 world-space.
1870 #+caption: Program to get the coordinates of ``feelers '' in
1871 #+caption: both world and UV-coordinates.
1872 #+name: feeler-coordinates
1873 #+begin_listing clojure
1874 #+BEGIN_SRC clojure
1875 (defn feeler-pixel-coords
1876 "Returns the coordinates of the feelers in pixel space in lists, one
1877 list for each triangle, ordered in the same way as (triangles) and
1878 (pixel-triangles)."
1879 [#^Geometry geo image]
1880 (map
1881 (fn [pixel-triangle]
1882 (filter
1883 (fn [coord]
1884 (inside-triangle? (->triangle pixel-triangle)
1885 (->vector3f coord)))
1886 (white-coordinates image (convex-bounds pixel-triangle))))
1887 (pixel-triangles geo image)))
1889 (defn feeler-world-coords
1890 "Returns the coordinates of the feelers in world space in lists, one
1891 list for each triangle, ordered in the same way as (triangles) and
1892 (pixel-triangles)."
1893 [#^Geometry geo image]
1894 (let [transforms
1895 (map #(triangles->affine-transform
1896 (->triangle %1) (->triangle %2))
1897 (pixel-triangles geo image)
1898 (triangles geo))]
1899 (map (fn [transform coords]
1900 (map #(.mult transform (->vector3f %)) coords))
1901 transforms (feeler-pixel-coords geo image))))
1902 #+END_SRC
1903 #+end_listing
1905 #+caption: Program to get the position of the base and tip of
1906 #+caption: each ``feeler''
1907 #+name: feeler-tips
1908 #+begin_listing clojure
1909 #+BEGIN_SRC clojure
1910 (defn feeler-origins
1911 "The world space coordinates of the root of each feeler."
1912 [#^Geometry geo image]
1913 (reduce concat (feeler-world-coords geo image)))
1915 (defn feeler-tips
1916 "The world space coordinates of the tip of each feeler."
1917 [#^Geometry geo image]
1918 (let [world-coords (feeler-world-coords geo image)
1919 normals
1920 (map
1921 (fn [triangle]
1922 (.calculateNormal triangle)
1923 (.clone (.getNormal triangle)))
1924 (map ->triangle (triangles geo)))]
1926 (mapcat (fn [origins normal]
1927 (map #(.add % normal) origins))
1928 world-coords normals)))
1930 (defn touch-topology
1931 [#^Geometry geo image]
1932 (collapse (reduce concat (feeler-pixel-coords geo image))))
1933 #+END_SRC
1934 #+end_listing
1936 *** Simulated Touch
1938 Now that the functions to construct feelers are complete,
1939 =touch-kernel= generates functions to be called from within a
1940 simulation that perform the necessary physics collisions to
1941 collect tactile data, and =touch!= recursively applies it to every
1942 node in the creature.
1944 #+caption: Efficient program to transform a ray from
1945 #+caption: one position to another.
1946 #+name: set-ray
1947 #+begin_listing clojure
1948 #+BEGIN_SRC clojure
1949 (defn set-ray [#^Ray ray #^Matrix4f transform
1950 #^Vector3f origin #^Vector3f tip]
1951 ;; Doing everything locally reduces garbage collection by enough to
1952 ;; be worth it.
1953 (.mult transform origin (.getOrigin ray))
1954 (.mult transform tip (.getDirection ray))
1955 (.subtractLocal (.getDirection ray) (.getOrigin ray))
1956 (.normalizeLocal (.getDirection ray)))
1957 #+END_SRC
1958 #+end_listing
1960 #+caption: This is the core of touch in =CORTEX= each feeler
1961 #+caption: follows the object it is bound to, reporting any
1962 #+caption: collisions that may happen.
1963 #+name: touch-kernel
1964 #+begin_listing clojure
1965 #+BEGIN_SRC clojure
1966 (defn touch-kernel
1967 "Constructs a function which will return tactile sensory data from
1968 'geo when called from inside a running simulation"
1969 [#^Geometry geo]
1970 (if-let
1971 [profile (tactile-sensor-profile geo)]
1972 (let [ray-reference-origins (feeler-origins geo profile)
1973 ray-reference-tips (feeler-tips geo profile)
1974 ray-length (tactile-scale geo)
1975 current-rays (map (fn [_] (Ray.)) ray-reference-origins)
1976 topology (touch-topology geo profile)
1977 correction (float (* ray-length -0.2))]
1978 ;; slight tolerance for very close collisions.
1979 (dorun
1980 (map (fn [origin tip]
1981 (.addLocal origin (.mult (.subtract tip origin)
1982 correction)))
1983 ray-reference-origins ray-reference-tips))
1984 (dorun (map #(.setLimit % ray-length) current-rays))
1985 (fn [node]
1986 (let [transform (.getWorldMatrix geo)]
1987 (dorun
1988 (map (fn [ray ref-origin ref-tip]
1989 (set-ray ray transform ref-origin ref-tip))
1990 current-rays ray-reference-origins
1991 ray-reference-tips))
1992 (vector
1993 topology
1994 (vec
1995 (for [ray current-rays]
1996 (do
1997 (let [results (CollisionResults.)]
1998 (.collideWith node ray results)
1999 (let [touch-objects
2000 (filter #(not (= geo (.getGeometry %)))
2001 results)
2002 limit (.getLimit ray)]
2003 [(if (empty? touch-objects)
2004 limit
2005 (let [response
2006 (apply min (map #(.getDistance %)
2007 touch-objects))]
2008 (FastMath/clamp
2009 (float
2010 (if (> response limit) (float 0.0)
2011 (+ response correction)))
2012 (float 0.0)
2013 limit)))
2014 limit])))))))))))
2015 #+END_SRC
2016 #+end_listing
2018 Armed with the =touch!= function, =CORTEX= becomes capable of
2019 giving creatures a sense of touch. A simple test is to create a
2020 cube that is outfitted with a uniform distrubition of touch
2021 sensors. It can feel the ground and any balls that it touches.
2023 #+caption: =CORTEX= interface for creating touch in a simulated
2024 #+caption: creature.
2025 #+name: touch
2026 #+begin_listing clojure
2027 #+BEGIN_SRC clojure
2028 (defn touch!
2029 "Endow the creature with the sense of touch. Returns a sequence of
2030 functions, one for each body part with a tactile-sensor-profile,
2031 each of which when called returns sensory data for that body part."
2032 [#^Node creature]
2033 (filter
2034 (comp not nil?)
2035 (map touch-kernel
2036 (filter #(isa? (class %) Geometry)
2037 (node-seq creature)))))
2038 #+END_SRC
2039 #+end_listing
2041 The tactile-sensor-profile image for the touch cube is a simple
2042 cross with a unifom distribution of touch sensors:
2044 #+caption: The touch profile for the touch-cube. Each pure white
2045 #+caption: pixel defines a touch sensitive feeler.
2046 #+name: touch-cube-uv-map
2047 #+ATTR_LaTeX: :width 7cm
2048 [[./images/touch-profile.png]]
2050 #+caption: The touch cube reacts to canonballs. The black, red,
2051 #+caption: and white cross on the right is a visual display of
2052 #+caption: the creature's touch. White means that it is feeling
2053 #+caption: something strongly, black is not feeling anything,
2054 #+caption: and gray is in-between. The cube can feel both the
2055 #+caption: floor and the ball. Notice that when the ball causes
2056 #+caption: the cube to tip, that the bottom face can still feel
2057 #+caption: part of the ground.
2058 #+name: touch-cube-uv-map
2059 #+ATTR_LaTeX: :width 15cm
2060 [[./images/touch-cube.png]]
2062 ** Proprioception is the sense that makes everything ``real''
2064 Close your eyes, and touch your nose with your right index finger.
2065 How did you do it? You could not see your hand, and neither your
2066 hand nor your nose could use the sense of touch to guide the path
2067 of your hand. There are no sound cues, and Taste and Smell
2068 certainly don't provide any help. You know where your hand is
2069 without your other senses because of Proprioception.
2071 Humans can sometimes loose this sense through viral infections or
2072 damage to the spinal cord or brain, and when they do, they loose
2073 the ability to control their own bodies without looking directly at
2074 the parts they want to move. In [[http://en.wikipedia.org/wiki/The_Man_Who_Mistook_His_Wife_for_a_Hat][The Man Who Mistook His Wife for a
2075 Hat]], a woman named Christina looses this sense and has to learn how
2076 to move by carefully watching her arms and legs. She describes
2077 proprioception as the "eyes of the body, the way the body sees
2078 itself".
2080 Proprioception in humans is mediated by [[http://en.wikipedia.org/wiki/Articular_capsule][joint capsules]], [[http://en.wikipedia.org/wiki/Muscle_spindle][muscle
2081 spindles]], and the [[http://en.wikipedia.org/wiki/Golgi_tendon_organ][Golgi tendon organs]]. These measure the relative
2082 positions of each body part by monitoring muscle strain and length.
2084 It's clear that this is a vital sense for fluid, graceful movement.
2085 It's also particularly easy to implement in jMonkeyEngine.
2087 My simulated proprioception calculates the relative angles of each
2088 joint from the rest position defined in the blender file. This
2089 simulates the muscle-spindles and joint capsules. I will deal with
2090 Golgi tendon organs, which calculate muscle strain, in the next
2091 section.
2093 *** Helper functions
2095 =absolute-angle= calculates the angle between two vectors,
2096 relative to a third axis vector. This angle is the number of
2097 radians you have to move counterclockwise around the axis vector
2098 to get from the first to the second vector. It is not commutative
2099 like a normal dot-product angle is.
2101 The purpose of these functions is to build a system of angle
2102 measurement that is biologically plausable.
2104 #+caption: Program to measure angles along a vector
2105 #+name: helpers
2106 #+begin_listing clojure
2107 #+BEGIN_SRC clojure
2108 (defn right-handed?
2109 "true iff the three vectors form a right handed coordinate
2110 system. The three vectors do not have to be normalized or
2111 orthogonal."
2112 [vec1 vec2 vec3]
2113 (pos? (.dot (.cross vec1 vec2) vec3)))
2115 (defn absolute-angle
2116 "The angle between 'vec1 and 'vec2 around 'axis. In the range
2117 [0 (* 2 Math/PI)]."
2118 [vec1 vec2 axis]
2119 (let [angle (.angleBetween vec1 vec2)]
2120 (if (right-handed? vec1 vec2 axis)
2121 angle (- (* 2 Math/PI) angle))))
2122 #+END_SRC
2123 #+end_listing
2125 *** Proprioception Kernel
2127 Given a joint, =proprioception-kernel= produces a function that
2128 calculates the Euler angles between the the objects the joint
2129 connects. The only tricky part here is making the angles relative
2130 to the joint's initial ``straightness''.
2132 #+caption: Program to return biologially reasonable proprioceptive
2133 #+caption: data for each joint.
2134 #+name: proprioception
2135 #+begin_listing clojure
2136 #+BEGIN_SRC clojure
2137 (defn proprioception-kernel
2138 "Returns a function which returns proprioceptive sensory data when
2139 called inside a running simulation."
2140 [#^Node parts #^Node joint]
2141 (let [[obj-a obj-b] (joint-targets parts joint)
2142 joint-rot (.getWorldRotation joint)
2143 x0 (.mult joint-rot Vector3f/UNIT_X)
2144 y0 (.mult joint-rot Vector3f/UNIT_Y)
2145 z0 (.mult joint-rot Vector3f/UNIT_Z)]
2146 (fn []
2147 (let [rot-a (.clone (.getWorldRotation obj-a))
2148 rot-b (.clone (.getWorldRotation obj-b))
2149 x (.mult rot-a x0)
2150 y (.mult rot-a y0)
2151 z (.mult rot-a z0)
2153 X (.mult rot-b x0)
2154 Y (.mult rot-b y0)
2155 Z (.mult rot-b z0)
2156 heading (Math/atan2 (.dot X z) (.dot X x))
2157 pitch (Math/atan2 (.dot X y) (.dot X x))
2159 ;; rotate x-vector back to origin
2160 reverse
2161 (doto (Quaternion.)
2162 (.fromAngleAxis
2163 (.angleBetween X x)
2164 (let [cross (.normalize (.cross X x))]
2165 (if (= 0 (.length cross)) y cross))))
2166 roll (absolute-angle (.mult reverse Y) y x)]
2167 [heading pitch roll]))))
2169 (defn proprioception!
2170 "Endow the creature with the sense of proprioception. Returns a
2171 sequence of functions, one for each child of the \"joints\" node in
2172 the creature, which each report proprioceptive information about
2173 that joint."
2174 [#^Node creature]
2175 ;; extract the body's joints
2176 (let [senses (map (partial proprioception-kernel creature)
2177 (joints creature))]
2178 (fn []
2179 (map #(%) senses))))
2180 #+END_SRC
2181 #+end_listing
2183 =proprioception!= maps =proprioception-kernel= across all the
2184 joints of the creature. It uses the same list of joints that
2185 =joints= uses. Proprioception is the easiest sense to implement in
2186 =CORTEX=, and it will play a crucial role when efficiently
2187 implementing empathy.
2189 #+caption: In the upper right corner, the three proprioceptive
2190 #+caption: angle measurements are displayed. Red is yaw, Green is
2191 #+caption: pitch, and White is roll.
2192 #+name: proprio
2193 #+ATTR_LaTeX: :width 11cm
2194 [[./images/proprio.png]]
2196 ** Muscles are both effectors and sensors
2198 Surprisingly enough, terrestrial creatures only move by using
2199 torque applied about their joints. There's not a single straight
2200 line of force in the human body at all! (A straight line of force
2201 would correspond to some sort of jet or rocket propulsion.)
2203 In humans, muscles are composed of muscle fibers which can contract
2204 to exert force. The muscle fibers which compose a muscle are
2205 partitioned into discrete groups which are each controlled by a
2206 single alpha motor neuron. A single alpha motor neuron might
2207 control as little as three or as many as one thousand muscle
2208 fibers. When the alpha motor neuron is engaged by the spinal cord,
2209 it activates all of the muscle fibers to which it is attached. The
2210 spinal cord generally engages the alpha motor neurons which control
2211 few muscle fibers before the motor neurons which control many
2212 muscle fibers. This recruitment strategy allows for precise
2213 movements at low strength. The collection of all motor neurons that
2214 control a muscle is called the motor pool. The brain essentially
2215 says "activate 30% of the motor pool" and the spinal cord recruits
2216 motor neurons until 30% are activated. Since the distribution of
2217 power among motor neurons is unequal and recruitment goes from
2218 weakest to strongest, the first 30% of the motor pool might be 5%
2219 of the strength of the muscle.
2221 My simulated muscles follow a similar design: Each muscle is
2222 defined by a 1-D array of numbers (the "motor pool"). Each entry in
2223 the array represents a motor neuron which controls a number of
2224 muscle fibers equal to the value of the entry. Each muscle has a
2225 scalar strength factor which determines the total force the muscle
2226 can exert when all motor neurons are activated. The effector
2227 function for a muscle takes a number to index into the motor pool,
2228 and then "activates" all the motor neurons whose index is lower or
2229 equal to the number. Each motor-neuron will apply force in
2230 proportion to its value in the array. Lower values cause less
2231 force. The lower values can be put at the "beginning" of the 1-D
2232 array to simulate the layout of actual human muscles, which are
2233 capable of more precise movements when exerting less force. Or, the
2234 motor pool can simulate more exotic recruitment strategies which do
2235 not correspond to human muscles.
2237 This 1D array is defined in an image file for ease of
2238 creation/visualization. Here is an example muscle profile image.
2240 #+caption: A muscle profile image that describes the strengths
2241 #+caption: of each motor neuron in a muscle. White is weakest
2242 #+caption: and dark red is strongest. This particular pattern
2243 #+caption: has weaker motor neurons at the beginning, just
2244 #+caption: like human muscle.
2245 #+name: muscle-recruit
2246 #+ATTR_LaTeX: :width 7cm
2247 [[./images/basic-muscle.png]]
2249 *** Muscle meta-data
2251 #+caption: Program to deal with loading muscle data from a blender
2252 #+caption: file's metadata.
2253 #+name: motor-pool
2254 #+begin_listing clojure
2255 #+BEGIN_SRC clojure
2256 (defn muscle-profile-image
2257 "Get the muscle-profile image from the node's blender meta-data."
2258 [#^Node muscle]
2259 (if-let [image (meta-data muscle "muscle")]
2260 (load-image image)))
2262 (defn muscle-strength
2263 "Return the strength of this muscle, or 1 if it is not defined."
2264 [#^Node muscle]
2265 (if-let [strength (meta-data muscle "strength")]
2266 strength 1))
2268 (defn motor-pool
2269 "Return a vector where each entry is the strength of the \"motor
2270 neuron\" at that part in the muscle."
2271 [#^Node muscle]
2272 (let [profile (muscle-profile-image muscle)]
2273 (vec
2274 (let [width (.getWidth profile)]
2275 (for [x (range width)]
2276 (- 255
2277 (bit-and
2278 0x0000FF
2279 (.getRGB profile x 0))))))))
2280 #+END_SRC
2281 #+end_listing
2283 Of note here is =motor-pool= which interprets the muscle-profile
2284 image in a way that allows me to use gradients between white and
2285 red, instead of shades of gray as I've been using for all the
2286 other senses. This is purely an aesthetic touch.
2288 *** Creating muscles
2290 #+caption: This is the core movement functoion in =CORTEX=, which
2291 #+caption: implements muscles that report on their activation.
2292 #+name: muscle-kernel
2293 #+begin_listing clojure
2294 #+BEGIN_SRC clojure
2295 (defn movement-kernel
2296 "Returns a function which when called with a integer value inside a
2297 running simulation will cause movement in the creature according
2298 to the muscle's position and strength profile. Each function
2299 returns the amount of force applied / max force."
2300 [#^Node creature #^Node muscle]
2301 (let [target (closest-node creature muscle)
2302 axis
2303 (.mult (.getWorldRotation muscle) Vector3f/UNIT_Y)
2304 strength (muscle-strength muscle)
2306 pool (motor-pool muscle)
2307 pool-integral (reductions + pool)
2308 forces
2309 (vec (map #(float (* strength (/ % (last pool-integral))))
2310 pool-integral))
2311 control (.getControl target RigidBodyControl)]
2312 ;;(println-repl (.getName target) axis)
2313 (fn [n]
2314 (let [pool-index (max 0 (min n (dec (count pool))))
2315 force (forces pool-index)]
2316 (.applyTorque control (.mult axis force))
2317 (float (/ force strength))))))
2319 (defn movement!
2320 "Endow the creature with the power of movement. Returns a sequence
2321 of functions, each of which accept an integer value and will
2322 activate their corresponding muscle."
2323 [#^Node creature]
2324 (for [muscle (muscles creature)]
2325 (movement-kernel creature muscle)))
2326 #+END_SRC
2327 #+end_listing
2330 =movement-kernel= creates a function that will move the nearest
2331 physical object to the muscle node. The muscle exerts a rotational
2332 force dependent on it's orientation to the object in the blender
2333 file. The function returned by =movement-kernel= is also a sense
2334 function: it returns the percent of the total muscle strength that
2335 is currently being employed. This is analogous to muscle tension
2336 in humans and completes the sense of proprioception begun in the
2337 last section.
2339 ** =CORTEX= brings complex creatures to life!
2341 The ultimate test of =CORTEX= is to create a creature with the full
2342 gamut of senses and put it though its paces.
2344 With all senses enabled, my right hand model looks like an
2345 intricate marionette hand with several strings for each finger:
2347 #+caption: View of the hand model with all sense nodes. You can see
2348 #+caption: the joint, muscle, ear, and eye nodess here.
2349 #+name: hand-nodes-1
2350 #+ATTR_LaTeX: :width 11cm
2351 [[./images/hand-with-all-senses2.png]]
2353 #+caption: An alternate view of the hand.
2354 #+name: hand-nodes-2
2355 #+ATTR_LaTeX: :width 15cm
2356 [[./images/hand-with-all-senses3.png]]
2358 With the hand fully rigged with senses, I can run it though a test
2359 that will test everything.
2361 #+caption: A full test of the hand with all senses. Note expecially
2362 #+caption: the interactions the hand has with itself: it feels
2363 #+caption: its own palm and fingers, and when it curls its fingers,
2364 #+caption: it sees them with its eye (which is located in the center
2365 #+caption: of the palm. The red block appears with a pure tone sound.
2366 #+caption: The hand then uses its muscles to launch the cube!
2367 #+name: integration
2368 #+ATTR_LaTeX: :width 16cm
2369 [[./images/integration.png]]
2371 ** =CORTEX= enables many possiblities for further research
2373 Often times, the hardest part of building a system involving
2374 creatures is dealing with physics and graphics. =CORTEX= removes
2375 much of this initial difficulty and leaves researchers free to
2376 directly pursue their ideas. I hope that even undergrads with a
2377 passing curiosity about simulated touch or creature evolution will
2378 be able to use cortex for experimentation. =CORTEX= is a completely
2379 simulated world, and far from being a disadvantage, its simulated
2380 nature enables you to create senses and creatures that would be
2381 impossible to make in the real world.
2383 While not by any means a complete list, here are some paths
2384 =CORTEX= is well suited to help you explore:
2386 - Empathy :: my empathy program leaves many areas for
2387 improvement, among which are using vision to infer
2388 proprioception and looking up sensory experience with imagined
2389 vision, touch, and sound.
2390 - Evolution :: Karl Sims created a rich environment for
2391 simulating the evolution of creatures on a connection
2392 machine. Today, this can be redone and expanded with =CORTEX=
2393 on an ordinary computer.
2394 - Exotic senses :: Cortex enables many fascinating senses that are
2395 not possible to build in the real world. For example,
2396 telekinesis is an interesting avenue to explore. You can also
2397 make a ``semantic'' sense which looks up metadata tags on
2398 objects in the environment the metadata tags might contain
2399 other sensory information.
2400 - Imagination via subworlds :: this would involve a creature with
2401 an effector which creates an entire new sub-simulation where
2402 the creature has direct control over placement/creation of
2403 objects via simulated telekinesis. The creature observes this
2404 sub-world through it's normal senses and uses its observations
2405 to make predictions about its top level world.
2406 - Simulated prescience :: step the simulation forward a few ticks,
2407 gather sensory data, then supply this data for the creature as
2408 one of its actual senses. The cost of prescience is slowing
2409 the simulation down by a factor proportional to however far
2410 you want the entities to see into the future. What happens
2411 when two evolved creatures that can each see into the future
2412 fight each other?
2413 - Swarm creatures :: Program a group of creatures that cooperate
2414 with each other. Because the creatures would be simulated, you
2415 could investigate computationally complex rules of behavior
2416 which still, from the group's point of view, would happen in
2417 ``real time''. Interactions could be as simple as cellular
2418 organisms communicating via flashing lights, or as complex as
2419 humanoids completing social tasks, etc.
2420 - =HACKER= for writing muscle-control programs :: Presented with
2421 low-level muscle control/ sense API, generate higher level
2422 programs for accomplishing various stated goals. Example goals
2423 might be "extend all your fingers" or "move your hand into the
2424 area with blue light" or "decrease the angle of this joint".
2425 It would be like Sussman's HACKER, except it would operate
2426 with much more data in a more realistic world. Start off with
2427 "calisthenics" to develop subroutines over the motor control
2428 API. This would be the "spinal chord" of a more intelligent
2429 creature. The low level programming code might be a turning
2430 machine that could develop programs to iterate over a "tape"
2431 where each entry in the tape could control recruitment of the
2432 fibers in a muscle.
2433 - Sense fusion :: There is much work to be done on sense
2434 integration -- building up a coherent picture of the world and
2435 the things in it with =CORTEX= as a base, you can explore
2436 concepts like self-organizing maps or cross modal clustering
2437 in ways that have never before been tried.
2438 - Inverse kinematics :: experiments in sense guided motor control
2439 are easy given =CORTEX='s support -- you can get right to the
2440 hard control problems without worrying about physics or
2441 senses.
2443 * Empathy in a simulated worm
2445 Here I develop a computational model of empathy, using =CORTEX= as a
2446 base. Empathy in this context is the ability to observe another
2447 creature and infer what sorts of sensations that creature is
2448 feeling. My empathy algorithm involves multiple phases. First is
2449 free-play, where the creature moves around and gains sensory
2450 experience. From this experience I construct a representation of the
2451 creature's sensory state space, which I call \Phi-space. Using
2452 \Phi-space, I construct an efficient function which takes the
2453 limited data that comes from observing another creature and enriches
2454 it full compliment of imagined sensory data. I can then use the
2455 imagined sensory data to recognize what the observed creature is
2456 doing and feeling, using straightforward embodied action predicates.
2457 This is all demonstrated with using a simple worm-like creature, and
2458 recognizing worm-actions based on limited data.
2460 #+caption: Here is the worm with which we will be working.
2461 #+caption: It is composed of 5 segments. Each segment has a
2462 #+caption: pair of extensor and flexor muscles. Each of the
2463 #+caption: worm's four joints is a hinge joint which allows
2464 #+caption: about 30 degrees of rotation to either side. Each segment
2465 #+caption: of the worm is touch-capable and has a uniform
2466 #+caption: distribution of touch sensors on each of its faces.
2467 #+caption: Each joint has a proprioceptive sense to detect
2468 #+caption: relative positions. The worm segments are all the
2469 #+caption: same except for the first one, which has a much
2470 #+caption: higher weight than the others to allow for easy
2471 #+caption: manual motor control.
2472 #+name: basic-worm-view
2473 #+ATTR_LaTeX: :width 10cm
2474 [[./images/basic-worm-view.png]]
2476 #+caption: Program for reading a worm from a blender file and
2477 #+caption: outfitting it with the senses of proprioception,
2478 #+caption: touch, and the ability to move, as specified in the
2479 #+caption: blender file.
2480 #+name: get-worm
2481 #+begin_listing clojure
2482 #+begin_src clojure
2483 (defn worm []
2484 (let [model (load-blender-model "Models/worm/worm.blend")]
2485 {:body (doto model (body!))
2486 :touch (touch! model)
2487 :proprioception (proprioception! model)
2488 :muscles (movement! model)}))
2489 #+end_src
2490 #+end_listing
2492 ** Embodiment factors action recognition into managable parts
2494 Using empathy, I divide the problem of action recognition into a
2495 recognition process expressed in the language of a full compliment
2496 of senses, and an imaganitive process that generates full sensory
2497 data from partial sensory data. Splitting the action recognition
2498 problem in this manner greatly reduces the total amount of work to
2499 recognize actions: The imaganitive process is mostly just matching
2500 previous experience, and the recognition process gets to use all
2501 the senses to directly describe any action.
2503 ** Action recognition is easy with a full gamut of senses
2505 Embodied representations using multiple senses such as touch,
2506 proprioception, and muscle tension turns out be be exceedingly
2507 efficient at describing body-centered actions. It is the ``right
2508 language for the job''. For example, it takes only around 5 lines
2509 of LISP code to describe the action of ``curling'' using embodied
2510 primitives. It takes about 10 lines to describe the seemingly
2511 complicated action of wiggling.
2513 The following action predicates each take a stream of sensory
2514 experience, observe however much of it they desire, and decide
2515 whether the worm is doing the action they describe. =curled?=
2516 relies on proprioception, =resting?= relies on touch, =wiggling?=
2517 relies on a fourier analysis of muscle contraction, and
2518 =grand-circle?= relies on touch and reuses =curled?= as a gaurd.
2520 #+caption: Program for detecting whether the worm is curled. This is the
2521 #+caption: simplest action predicate, because it only uses the last frame
2522 #+caption: of sensory experience, and only uses proprioceptive data. Even
2523 #+caption: this simple predicate, however, is automatically frame
2524 #+caption: independent and ignores vermopomorphic differences such as
2525 #+caption: worm textures and colors.
2526 #+name: curled
2527 #+begin_listing clojure
2528 #+begin_src clojure
2529 (defn curled?
2530 "Is the worm curled up?"
2531 [experiences]
2532 (every?
2533 (fn [[_ _ bend]]
2534 (> (Math/sin bend) 0.64))
2535 (:proprioception (peek experiences))))
2536 #+end_src
2537 #+end_listing
2539 #+caption: Program for summarizing the touch information in a patch
2540 #+caption: of skin.
2541 #+name: touch-summary
2542 #+begin_listing clojure
2543 #+begin_src clojure
2544 (defn contact
2545 "Determine how much contact a particular worm segment has with
2546 other objects. Returns a value between 0 and 1, where 1 is full
2547 contact and 0 is no contact."
2548 [touch-region [coords contact :as touch]]
2549 (-> (zipmap coords contact)
2550 (select-keys touch-region)
2551 (vals)
2552 (#(map first %))
2553 (average)
2554 (* 10)
2555 (- 1)
2556 (Math/abs)))
2557 #+end_src
2558 #+end_listing
2561 #+caption: Program for detecting whether the worm is at rest. This program
2562 #+caption: uses a summary of the tactile information from the underbelly
2563 #+caption: of the worm, and is only true if every segment is touching the
2564 #+caption: floor. Note that this function contains no references to
2565 #+caption: proprioction at all.
2566 #+name: resting
2567 #+begin_listing clojure
2568 #+begin_src clojure
2569 (def worm-segment-bottom (rect-region [8 15] [14 22]))
2571 (defn resting?
2572 "Is the worm resting on the ground?"
2573 [experiences]
2574 (every?
2575 (fn [touch-data]
2576 (< 0.9 (contact worm-segment-bottom touch-data)))
2577 (:touch (peek experiences))))
2578 #+end_src
2579 #+end_listing
2581 #+caption: Program for detecting whether the worm is curled up into a
2582 #+caption: full circle. Here the embodied approach begins to shine, as
2583 #+caption: I am able to both use a previous action predicate (=curled?=)
2584 #+caption: as well as the direct tactile experience of the head and tail.
2585 #+name: grand-circle
2586 #+begin_listing clojure
2587 #+begin_src clojure
2588 (def worm-segment-bottom-tip (rect-region [15 15] [22 22]))
2590 (def worm-segment-top-tip (rect-region [0 15] [7 22]))
2592 (defn grand-circle?
2593 "Does the worm form a majestic circle (one end touching the other)?"
2594 [experiences]
2595 (and (curled? experiences)
2596 (let [worm-touch (:touch (peek experiences))
2597 tail-touch (worm-touch 0)
2598 head-touch (worm-touch 4)]
2599 (and (< 0.55 (contact worm-segment-bottom-tip tail-touch))
2600 (< 0.55 (contact worm-segment-top-tip head-touch))))))
2601 #+end_src
2602 #+end_listing
2605 #+caption: Program for detecting whether the worm has been wiggling for
2606 #+caption: the last few frames. It uses a fourier analysis of the muscle
2607 #+caption: contractions of the worm's tail to determine wiggling. This is
2608 #+caption: signigicant because there is no particular frame that clearly
2609 #+caption: indicates that the worm is wiggling --- only when multiple frames
2610 #+caption: are analyzed together is the wiggling revealed. Defining
2611 #+caption: wiggling this way also gives the worm an opportunity to learn
2612 #+caption: and recognize ``frustrated wiggling'', where the worm tries to
2613 #+caption: wiggle but can't. Frustrated wiggling is very visually different
2614 #+caption: from actual wiggling, but this definition gives it to us for free.
2615 #+name: wiggling
2616 #+begin_listing clojure
2617 #+begin_src clojure
2618 (defn fft [nums]
2619 (map
2620 #(.getReal %)
2621 (.transform
2622 (FastFourierTransformer. DftNormalization/STANDARD)
2623 (double-array nums) TransformType/FORWARD)))
2625 (def indexed (partial map-indexed vector))
2627 (defn max-indexed [s]
2628 (first (sort-by (comp - second) (indexed s))))
2630 (defn wiggling?
2631 "Is the worm wiggling?"
2632 [experiences]
2633 (let [analysis-interval 0x40]
2634 (when (> (count experiences) analysis-interval)
2635 (let [a-flex 3
2636 a-ex 2
2637 muscle-activity
2638 (map :muscle (vector:last-n experiences analysis-interval))
2639 base-activity
2640 (map #(- (% a-flex) (% a-ex)) muscle-activity)]
2641 (= 2
2642 (first
2643 (max-indexed
2644 (map #(Math/abs %)
2645 (take 20 (fft base-activity))))))))))
2646 #+end_src
2647 #+end_listing
2649 With these action predicates, I can now recognize the actions of
2650 the worm while it is moving under my control and I have access to
2651 all the worm's senses.
2653 #+caption: Use the action predicates defined earlier to report on
2654 #+caption: what the worm is doing while in simulation.
2655 #+name: report-worm-activity
2656 #+begin_listing clojure
2657 #+begin_src clojure
2658 (defn debug-experience
2659 [experiences text]
2660 (cond
2661 (grand-circle? experiences) (.setText text "Grand Circle")
2662 (curled? experiences) (.setText text "Curled")
2663 (wiggling? experiences) (.setText text "Wiggling")
2664 (resting? experiences) (.setText text "Resting")))
2665 #+end_src
2666 #+end_listing
2668 #+caption: Using =debug-experience=, the body-centered predicates
2669 #+caption: work together to classify the behaviour of the worm.
2670 #+caption: the predicates are operating with access to the worm's
2671 #+caption: full sensory data.
2672 #+name: basic-worm-view
2673 #+ATTR_LaTeX: :width 10cm
2674 [[./images/worm-identify-init.png]]
2676 These action predicates satisfy the recognition requirement of an
2677 empathic recognition system. There is power in the simplicity of
2678 the action predicates. They describe their actions without getting
2679 confused in visual details of the worm. Each one is frame
2680 independent, but more than that, they are each indepent of
2681 irrelevant visual details of the worm and the environment. They
2682 will work regardless of whether the worm is a different color or
2683 hevaily textured, or if the environment has strange lighting.
2685 The trick now is to make the action predicates work even when the
2686 sensory data on which they depend is absent. If I can do that, then
2687 I will have gained much,
2689 ** \Phi-space describes the worm's experiences
2691 As a first step towards building empathy, I need to gather all of
2692 the worm's experiences during free play. I use a simple vector to
2693 store all the experiences.
2695 Each element of the experience vector exists in the vast space of
2696 all possible worm-experiences. Most of this vast space is actually
2697 unreachable due to physical constraints of the worm's body. For
2698 example, the worm's segments are connected by hinge joints that put
2699 a practical limit on the worm's range of motions without limiting
2700 its degrees of freedom. Some groupings of senses are impossible;
2701 the worm can not be bent into a circle so that its ends are
2702 touching and at the same time not also experience the sensation of
2703 touching itself.
2705 As the worm moves around during free play and its experience vector
2706 grows larger, the vector begins to define a subspace which is all
2707 the sensations the worm can practicaly experience during normal
2708 operation. I call this subspace \Phi-space, short for
2709 physical-space. The experience vector defines a path through
2710 \Phi-space. This path has interesting properties that all derive
2711 from physical embodiment. The proprioceptive components are
2712 completely smooth, because in order for the worm to move from one
2713 position to another, it must pass through the intermediate
2714 positions. The path invariably forms loops as actions are repeated.
2715 Finally and most importantly, proprioception actually gives very
2716 strong inference about the other senses. For example, when the worm
2717 is flat, you can infer that it is touching the ground and that its
2718 muscles are not active, because if the muscles were active, the
2719 worm would be moving and would not be perfectly flat. In order to
2720 stay flat, the worm has to be touching the ground, or it would
2721 again be moving out of the flat position due to gravity. If the
2722 worm is positioned in such a way that it interacts with itself,
2723 then it is very likely to be feeling the same tactile feelings as
2724 the last time it was in that position, because it has the same body
2725 as then. If you observe multiple frames of proprioceptive data,
2726 then you can become increasingly confident about the exact
2727 activations of the worm's muscles, because it generally takes a
2728 unique combination of muscle contractions to transform the worm's
2729 body along a specific path through \Phi-space.
2731 There is a simple way of taking \Phi-space and the total ordering
2732 provided by an experience vector and reliably infering the rest of
2733 the senses.
2735 ** Empathy is the process of tracing though \Phi-space
2737 Here is the core of a basic empathy algorithm, starting with an
2738 experience vector:
2740 First, group the experiences into tiered proprioceptive bins. I use
2741 powers of 10 and 3 bins, and the smallest bin has an approximate
2742 size of 0.001 radians in all proprioceptive dimensions.
2744 Then, given a sequence of proprioceptive input, generate a set of
2745 matching experience records for each input, using the tiered
2746 proprioceptive bins.
2748 Finally, to infer sensory data, select the longest consective chain
2749 of experiences. Conecutive experience means that the experiences
2750 appear next to each other in the experience vector.
2752 This algorithm has three advantages:
2754 1. It's simple
2756 3. It's very fast -- retrieving possible interpretations takes
2757 constant time. Tracing through chains of interpretations takes
2758 time proportional to the average number of experiences in a
2759 proprioceptive bin. Redundant experiences in \Phi-space can be
2760 merged to save computation.
2762 2. It protects from wrong interpretations of transient ambiguous
2763 proprioceptive data. For example, if the worm is flat for just
2764 an instant, this flattness will not be interpreted as implying
2765 that the worm has its muscles relaxed, since the flattness is
2766 part of a longer chain which includes a distinct pattern of
2767 muscle activation. Markov chains or other memoryless statistical
2768 models that operate on individual frames may very well make this
2769 mistake.
2771 #+caption: Program to convert an experience vector into a
2772 #+caption: proprioceptively binned lookup function.
2773 #+name: bin
2774 #+begin_listing clojure
2775 #+begin_src clojure
2776 (defn bin [digits]
2777 (fn [angles]
2778 (->> angles
2779 (flatten)
2780 (map (juxt #(Math/sin %) #(Math/cos %)))
2781 (flatten)
2782 (mapv #(Math/round (* % (Math/pow 10 (dec digits))))))))
2784 (defn gen-phi-scan
2785 "Nearest-neighbors with binning. Only returns a result if
2786 the propriceptive data is within 10% of a previously recorded
2787 result in all dimensions."
2788 [phi-space]
2789 (let [bin-keys (map bin [3 2 1])
2790 bin-maps
2791 (map (fn [bin-key]
2792 (group-by
2793 (comp bin-key :proprioception phi-space)
2794 (range (count phi-space)))) bin-keys)
2795 lookups (map (fn [bin-key bin-map]
2796 (fn [proprio] (bin-map (bin-key proprio))))
2797 bin-keys bin-maps)]
2798 (fn lookup [proprio-data]
2799 (set (some #(% proprio-data) lookups)))))
2800 #+end_src
2801 #+end_listing
2803 #+caption: =longest-thread= finds the longest path of consecutive
2804 #+caption: experiences to explain proprioceptive worm data.
2805 #+name: phi-space-history-scan
2806 #+ATTR_LaTeX: :width 10cm
2807 [[./images/aurellem-gray.png]]
2809 =longest-thread= infers sensory data by stitching together pieces
2810 from previous experience. It prefers longer chains of previous
2811 experience to shorter ones. For example, during training the worm
2812 might rest on the ground for one second before it performs its
2813 excercises. If during recognition the worm rests on the ground for
2814 five seconds, =longest-thread= will accomodate this five second
2815 rest period by looping the one second rest chain five times.
2817 =longest-thread= takes time proportinal to the average number of
2818 entries in a proprioceptive bin, because for each element in the
2819 starting bin it performes a series of set lookups in the preceeding
2820 bins. If the total history is limited, then this is only a constant
2821 multiple times the number of entries in the starting bin. This
2822 analysis also applies even if the action requires multiple longest
2823 chains -- it's still the average number of entries in a
2824 proprioceptive bin times the desired chain length. Because
2825 =longest-thread= is so efficient and simple, I can interpret
2826 worm-actions in real time.
2828 #+caption: Program to calculate empathy by tracing though \Phi-space
2829 #+caption: and finding the longest (ie. most coherent) interpretation
2830 #+caption: of the data.
2831 #+name: longest-thread
2832 #+begin_listing clojure
2833 #+begin_src clojure
2834 (defn longest-thread
2835 "Find the longest thread from phi-index-sets. The index sets should
2836 be ordered from most recent to least recent."
2837 [phi-index-sets]
2838 (loop [result '()
2839 [thread-bases & remaining :as phi-index-sets] phi-index-sets]
2840 (if (empty? phi-index-sets)
2841 (vec result)
2842 (let [threads
2843 (for [thread-base thread-bases]
2844 (loop [thread (list thread-base)
2845 remaining remaining]
2846 (let [next-index (dec (first thread))]
2847 (cond (empty? remaining) thread
2848 (contains? (first remaining) next-index)
2849 (recur
2850 (cons next-index thread) (rest remaining))
2851 :else thread))))
2852 longest-thread
2853 (reduce (fn [thread-a thread-b]
2854 (if (> (count thread-a) (count thread-b))
2855 thread-a thread-b))
2856 '(nil)
2857 threads)]
2858 (recur (concat longest-thread result)
2859 (drop (count longest-thread) phi-index-sets))))))
2860 #+end_src
2861 #+end_listing
2863 There is one final piece, which is to replace missing sensory data
2864 with a best-guess estimate. While I could fill in missing data by
2865 using a gradient over the closest known sensory data points,
2866 averages can be misleading. It is certainly possible to create an
2867 impossible sensory state by averaging two possible sensory states.
2868 Therefore, I simply replicate the most recent sensory experience to
2869 fill in the gaps.
2871 #+caption: Fill in blanks in sensory experience by replicating the most
2872 #+caption: recent experience.
2873 #+name: infer-nils
2874 #+begin_listing clojure
2875 #+begin_src clojure
2876 (defn infer-nils
2877 "Replace nils with the next available non-nil element in the
2878 sequence, or barring that, 0."
2879 [s]
2880 (loop [i (dec (count s))
2881 v (transient s)]
2882 (if (zero? i) (persistent! v)
2883 (if-let [cur (v i)]
2884 (if (get v (dec i) 0)
2885 (recur (dec i) v)
2886 (recur (dec i) (assoc! v (dec i) cur)))
2887 (recur i (assoc! v i 0))))))
2888 #+end_src
2889 #+end_listing
2891 ** Efficient action recognition with =EMPATH=
2893 To use =EMPATH= with the worm, I first need to gather a set of
2894 experiences from the worm that includes the actions I want to
2895 recognize. The =generate-phi-space= program (listing
2896 \ref{generate-phi-space} runs the worm through a series of
2897 exercices and gatheres those experiences into a vector. The
2898 =do-all-the-things= program is a routine expressed in a simple
2899 muscle contraction script language for automated worm control. It
2900 causes the worm to rest, curl, and wiggle over about 700 frames
2901 (approx. 11 seconds).
2903 #+caption: Program to gather the worm's experiences into a vector for
2904 #+caption: further processing. The =motor-control-program= line uses
2905 #+caption: a motor control script that causes the worm to execute a series
2906 #+caption: of ``exercices'' that include all the action predicates.
2907 #+name: generate-phi-space
2908 #+begin_listing clojure
2909 #+begin_src clojure
2910 (def do-all-the-things
2911 (concat
2912 curl-script
2913 [[300 :d-ex 40]
2914 [320 :d-ex 0]]
2915 (shift-script 280 (take 16 wiggle-script))))
2917 (defn generate-phi-space []
2918 (let [experiences (atom [])]
2919 (run-world
2920 (apply-map
2921 worm-world
2922 (merge
2923 (worm-world-defaults)
2924 {:end-frame 700
2925 :motor-control
2926 (motor-control-program worm-muscle-labels do-all-the-things)
2927 :experiences experiences})))
2928 @experiences))
2929 #+end_src
2930 #+end_listing
2932 #+caption: Use longest thread and a phi-space generated from a short
2933 #+caption: exercise routine to interpret actions during free play.
2934 #+name: empathy-debug
2935 #+begin_listing clojure
2936 #+begin_src clojure
2937 (defn init []
2938 (def phi-space (generate-phi-space))
2939 (def phi-scan (gen-phi-scan phi-space)))
2941 (defn empathy-demonstration []
2942 (let [proprio (atom ())]
2943 (fn
2944 [experiences text]
2945 (let [phi-indices (phi-scan (:proprioception (peek experiences)))]
2946 (swap! proprio (partial cons phi-indices))
2947 (let [exp-thread (longest-thread (take 300 @proprio))
2948 empathy (mapv phi-space (infer-nils exp-thread))]
2949 (println-repl (vector:last-n exp-thread 22))
2950 (cond
2951 (grand-circle? empathy) (.setText text "Grand Circle")
2952 (curled? empathy) (.setText text "Curled")
2953 (wiggling? empathy) (.setText text "Wiggling")
2954 (resting? empathy) (.setText text "Resting")
2955 :else (.setText text "Unknown")))))))
2957 (defn empathy-experiment [record]
2958 (.start (worm-world :experience-watch (debug-experience-phi)
2959 :record record :worm worm*)))
2960 #+end_src
2961 #+end_listing
2963 The result of running =empathy-experiment= is that the system is
2964 generally able to interpret worm actions using the action-predicates
2965 on simulated sensory data just as well as with actual data. Figure
2966 \ref{empathy-debug-image} was generated using =empathy-experiment=:
2968 #+caption: From only proprioceptive data, =EMPATH= was able to infer
2969 #+caption: the complete sensory experience and classify four poses
2970 #+caption: (The last panel shows a composite image of \emph{wriggling},
2971 #+caption: a dynamic pose.)
2972 #+name: empathy-debug-image
2973 #+ATTR_LaTeX: :width 10cm :placement [H]
2974 [[./images/empathy-1.png]]
2976 One way to measure the performance of =EMPATH= is to compare the
2977 sutiability of the imagined sense experience to trigger the same
2978 action predicates as the real sensory experience.
2980 #+caption: Determine how closely empathy approximates actual
2981 #+caption: sensory data.
2982 #+name: test-empathy-accuracy
2983 #+begin_listing clojure
2984 #+begin_src clojure
2985 (def worm-action-label
2986 (juxt grand-circle? curled? wiggling?))
2988 (defn compare-empathy-with-baseline [matches]
2989 (let [proprio (atom ())]
2990 (fn
2991 [experiences text]
2992 (let [phi-indices (phi-scan (:proprioception (peek experiences)))]
2993 (swap! proprio (partial cons phi-indices))
2994 (let [exp-thread (longest-thread (take 300 @proprio))
2995 empathy (mapv phi-space (infer-nils exp-thread))
2996 experience-matches-empathy
2997 (= (worm-action-label experiences)
2998 (worm-action-label empathy))]
2999 (println-repl experience-matches-empathy)
3000 (swap! matches #(conj % experience-matches-empathy)))))))
3002 (defn accuracy [v]
3003 (float (/ (count (filter true? v)) (count v))))
3005 (defn test-empathy-accuracy []
3006 (let [res (atom [])]
3007 (run-world
3008 (worm-world :experience-watch
3009 (compare-empathy-with-baseline res)
3010 :worm worm*))
3011 (accuracy @res)))
3012 #+end_src
3013 #+end_listing
3015 Running =test-empathy-accuracy= using the very short exercise
3016 program defined in listing \ref{generate-phi-space}, and then doing
3017 a similar pattern of activity manually yeilds an accuracy of around
3018 73%. This is based on very limited worm experience. By training the
3019 worm for longer, the accuracy dramatically improves.
3021 #+caption: Program to generate \Phi-space using manual training.
3022 #+name: manual-phi-space
3023 #+begin_listing clojure
3024 #+begin_src clojure
3025 (defn init-interactive []
3026 (def phi-space
3027 (let [experiences (atom [])]
3028 (run-world
3029 (apply-map
3030 worm-world
3031 (merge
3032 (worm-world-defaults)
3033 {:experiences experiences})))
3034 @experiences))
3035 (def phi-scan (gen-phi-scan phi-space)))
3036 #+end_src
3037 #+end_listing
3039 After about 1 minute of manual training, I was able to achieve 95%
3040 accuracy on manual testing of the worm using =init-interactive= and
3041 =test-empathy-accuracy=. The majority of errors are near the
3042 boundaries of transitioning from one type of action to another.
3043 During these transitions the exact label for the action is more open
3044 to interpretation, and dissaggrement between empathy and experience
3045 is more excusable.
3047 ** Digression: bootstrapping touch using free exploration
3049 In the previous section I showed how to compute actions in terms of
3050 body-centered predicates which relied averate touch activation of
3051 pre-defined regions of the worm's skin. What if, instead of recieving
3052 touch pre-grouped into the six faces of each worm segment, the true
3053 topology of the worm's skin was unknown? This is more similiar to how
3054 a nerve fiber bundle might be arranged. While two fibers that are
3055 close in a nerve bundle /might/ correspond to two touch sensors that
3056 are close together on the skin, the process of taking a complicated
3057 surface and forcing it into essentially a circle requires some cuts
3058 and rerragenments.
3060 In this section I show how to automatically learn the skin-topology of
3061 a worm segment by free exploration. As the worm rolls around on the
3062 floor, large sections of its surface get activated. If the worm has
3063 stopped moving, then whatever region of skin that is touching the
3064 floor is probably an important region, and should be recorded.
3066 #+caption: Program to detect whether the worm is in a resting state
3067 #+caption: with one face touching the floor.
3068 #+name: pure-touch
3069 #+begin_listing clojure
3070 #+begin_src clojure
3071 (def full-contact [(float 0.0) (float 0.1)])
3073 (defn pure-touch?
3074 "This is worm specific code to determine if a large region of touch
3075 sensors is either all on or all off."
3076 [[coords touch :as touch-data]]
3077 (= (set (map first touch)) (set full-contact)))
3078 #+end_src
3079 #+end_listing
3081 After collecting these important regions, there will many nearly
3082 similiar touch regions. While for some purposes the subtle
3083 differences between these regions will be important, for my
3084 purposes I colapse them into mostly non-overlapping sets using
3085 =remove-similiar= in listing \ref{remove-similiar}
3087 #+caption: Program to take a lits of set of points and ``collapse them''
3088 #+caption: so that the remaining sets in the list are siginificantly
3089 #+caption: different from each other. Prefer smaller sets to larger ones.
3090 #+name: remove-similiar
3091 #+begin_listing clojure
3092 #+begin_src clojure
3093 (defn remove-similar
3094 [coll]
3095 (loop [result () coll (sort-by (comp - count) coll)]
3096 (if (empty? coll) result
3097 (let [[x & xs] coll
3098 c (count x)]
3099 (if (some
3100 (fn [other-set]
3101 (let [oc (count other-set)]
3102 (< (- (count (union other-set x)) c) (* oc 0.1))))
3103 xs)
3104 (recur result xs)
3105 (recur (cons x result) xs))))))
3106 #+end_src
3107 #+end_listing
3109 Actually running this simulation is easy given =CORTEX='s facilities.
3111 #+caption: Collect experiences while the worm moves around. Filter the touch
3112 #+caption: sensations by stable ones, collapse similiar ones together,
3113 #+caption: and report the regions learned.
3114 #+name: learn-touch
3115 #+begin_listing clojure
3116 #+begin_src clojure
3117 (defn learn-touch-regions []
3118 (let [experiences (atom [])
3119 world (apply-map
3120 worm-world
3121 (assoc (worm-segment-defaults)
3122 :experiences experiences))]
3123 (run-world world)
3124 (->>
3125 @experiences
3126 (drop 175)
3127 ;; access the single segment's touch data
3128 (map (comp first :touch))
3129 ;; only deal with "pure" touch data to determine surfaces
3130 (filter pure-touch?)
3131 ;; associate coordinates with touch values
3132 (map (partial apply zipmap))
3133 ;; select those regions where contact is being made
3134 (map (partial group-by second))
3135 (map #(get % full-contact))
3136 (map (partial map first))
3137 ;; remove redundant/subset regions
3138 (map set)
3139 remove-similar)))
3141 (defn learn-and-view-touch-regions []
3142 (map view-touch-region
3143 (learn-touch-regions)))
3144 #+end_src
3145 #+end_listing
3147 The only thing remining to define is the particular motion the worm
3148 must take. I accomplish this with a simple motor control program.
3150 #+caption: Motor control program for making the worm roll on the ground.
3151 #+caption: This could also be replaced with random motion.
3152 #+name: worm-roll
3153 #+begin_listing clojure
3154 #+begin_src clojure
3155 (defn touch-kinesthetics []
3156 [[170 :lift-1 40]
3157 [190 :lift-1 19]
3158 [206 :lift-1 0]
3160 [400 :lift-2 40]
3161 [410 :lift-2 0]
3163 [570 :lift-2 40]
3164 [590 :lift-2 21]
3165 [606 :lift-2 0]
3167 [800 :lift-1 30]
3168 [809 :lift-1 0]
3170 [900 :roll-2 40]
3171 [905 :roll-2 20]
3172 [910 :roll-2 0]
3174 [1000 :roll-2 40]
3175 [1005 :roll-2 20]
3176 [1010 :roll-2 0]
3178 [1100 :roll-2 40]
3179 [1105 :roll-2 20]
3180 [1110 :roll-2 0]
3181 ])
3182 #+end_src
3183 #+end_listing
3186 #+caption: The small worm rolls around on the floor, driven
3187 #+caption: by the motor control program in listing \ref{worm-roll}.
3188 #+name: worm-roll
3189 #+ATTR_LaTeX: :width 12cm
3190 [[./images/worm-roll.png]]
3193 #+caption: After completing its adventures, the worm now knows
3194 #+caption: how its touch sensors are arranged along its skin. These
3195 #+caption: are the regions that were deemed important by
3196 #+caption: =learn-touch-regions=. Note that the worm has discovered
3197 #+caption: that it has six sides.
3198 #+name: worm-touch-map
3199 #+ATTR_LaTeX: :width 12cm
3200 [[./images/touch-learn.png]]
3202 While simple, =learn-touch-regions= exploits regularities in both
3203 the worm's physiology and the worm's environment to correctly
3204 deduce that the worm has six sides. Note that =learn-touch-regions=
3205 would work just as well even if the worm's touch sense data were
3206 completely scrambled. The cross shape is just for convienence. This
3207 example justifies the use of pre-defined touch regions in =EMPATH=.
3209 * Contributions
3211 In this thesis you have seen the =CORTEX= system, a complete
3212 environment for creating simulated creatures. You have seen how to
3213 implement five senses including touch, proprioception, hearing,
3214 vision, and muscle tension. You have seen how to create new creatues
3215 using blender, a 3D modeling tool. I hope that =CORTEX= will be
3216 useful in further research projects. To this end I have included the
3217 full source to =CORTEX= along with a large suite of tests and
3218 examples. I have also created a user guide for =CORTEX= which is
3219 inculded in an appendix to this thesis.
3221 You have also seen how I used =CORTEX= as a platform to attach the
3222 /action recognition/ problem, which is the problem of recognizing
3223 actions in video. You saw a simple system called =EMPATH= which
3224 ientifies actions by first describing actions in a body-centerd,
3225 rich sense language, then infering a full range of sensory
3226 experience from limited data using previous experience gained from
3227 free play.
3229 As a minor digression, you also saw how I used =CORTEX= to enable a
3230 tiny worm to discover the topology of its skin simply by rolling on
3231 the ground.
3233 In conclusion, the main contributions of this thesis are:
3235 - =CORTEX=, a system for creating simulated creatures with rich
3236 senses.
3237 - =EMPATH=, a program for recognizing actions by imagining sensory
3238 experience.
3240 # An anatomical joke:
3241 # - Training
3242 # - Skeletal imitation
3243 # - Sensory fleshing-out
3244 # - Classification
3245 #+BEGIN_LaTeX
3246 \appendix
3247 #+END_LaTeX
3248 * Appendix: =CORTEX= User Guide
3250 Those who write a thesis should endeavor to make their code not only
3251 accessable, but actually useable, as a way to pay back the community
3252 that made the thesis possible in the first place. This thesis would
3253 not be possible without Free Software such as jMonkeyEngine3,
3254 Blender, clojure, emacs, ffmpeg, and many other tools. That is why I
3255 have included this user guide, in the hope that someone else might
3256 find =CORTEX= useful.
3258 ** Obtaining =CORTEX=
3260 You can get cortex from its mercurial repository at
3261 http://hg.bortreb.com/cortex. You may also download =CORTEX=
3262 releases at http://aurellem.org/cortex/releases/. As a condition of
3263 making this thesis, I have also provided Professor Winston the
3264 =CORTEX= source, and he knows how to run the demos and get started.
3265 You may also email me at =cortex@aurellem.org= and I may help where
3266 I can.
3268 ** Running =CORTEX=
3270 =CORTEX= comes with README and INSTALL files that will guide you
3271 through installation and running the test suite. In particular you
3272 should look at test =cortex.test= which contains test suites that
3273 run through all senses and multiple creatures.
3275 ** Creating creatures
3277 Creatures are created using /Blender/, a free 3D modeling program.
3278 You will need Blender version 2.6 when using the =CORTEX= included
3279 in this thesis. You create a =CORTEX= creature in a similiar manner
3280 to modeling anything in Blender, except that you also create
3281 several trees of empty nodes which define the creature's senses.
3283 *** Mass
3285 To give an object mass in =CORTEX=, add a ``mass'' metadata label
3286 to the object with the mass in jMonkeyEngine units. Note that
3287 setting the mass to 0 causes the object to be immovable.
3289 *** Joints
3291 Joints are created by creating an empty node named =joints= and
3292 then creating any number of empty child nodes to represent your
3293 creature's joints. The joint will automatically connect the
3294 closest two physical objects. It will help to set the empty node's
3295 display mode to ``Arrows'' so that you can clearly see the
3296 direction of the axes.
3298 Joint nodes should have the following metadata under the ``joint''
3299 label:
3301 #+BEGIN_SRC clojure
3302 ;; ONE OF the following, under the label "joint":
3303 {:type :point}
3305 ;; OR
3307 {:type :hinge
3308 :limit [<limit-low> <limit-high>]
3309 :axis (Vector3f. <x> <y> <z>)}
3310 ;;(:axis defaults to (Vector3f. 1 0 0) if not provided for hinge joints)
3312 ;; OR
3314 {:type :cone
3315 :limit-xz <lim-xz>
3316 :limit-xy <lim-xy>
3317 :twist <lim-twist>} ;(use XZY rotation mode in blender!)
3318 #+END_SRC
3320 *** Eyes
3322 Eyes are created by creating an empty node named =eyes= and then
3323 creating any number of empty child nodes to represent your
3324 creature's eyes.
3326 Eye nodes should have the following metadata under the ``eye''
3327 label:
3329 #+BEGIN_SRC clojure
3330 {:red <red-retina-definition>
3331 :blue <blue-retina-definition>
3332 :green <green-retina-definition>
3333 :all <all-retina-definition>
3334 (<0xrrggbb> <custom-retina-image>)...
3336 #+END_SRC
3338 Any of the color channels may be omitted. You may also include
3339 your own color selectors, and in fact :red is equivalent to
3340 0xFF0000 and so forth. The eye will be placed at the same position
3341 as the empty node and will bind to the neatest physical object.
3342 The eye will point outward from the X-axis of the node, and ``up''
3343 will be in the direction of the X-axis of the node. It will help
3344 to set the empty node's display mode to ``Arrows'' so that you can
3345 clearly see the direction of the axes.
3347 Each retina file should contain white pixels whever you want to be
3348 sensitive to your chosen color. If you want the entire field of
3349 view, specify :all of 0xFFFFFF and a retinal map that is entirely
3350 white.
3352 Here is a sample retinal map:
3354 #+caption: An example retinal profile image. White pixels are
3355 #+caption: photo-sensitive elements. The distribution of white
3356 #+caption: pixels is denser in the middle and falls off at the
3357 #+caption: edges and is inspired by the human retina.
3358 #+name: retina
3359 #+ATTR_LaTeX: :width 7cm :placement [H]
3360 [[./images/retina-small.png]]
3362 *** Hearing
3364 Ears are created by creating an empty node named =ears= and then
3365 creating any number of empty child nodes to represent your
3366 creature's ears.
3368 Ear nodes do not require any metadata.
3370 The ear will bind to and follow the closest physical node.
3372 *** Touch
3374 Touch is handled similarly to mass. To make a particular object
3375 touch sensitive, add metadata of the following form under the
3376 object's ``touch'' metadata field:
3378 #+BEGIN_EXAMPLE
3379 <touch-UV-map-file-name>
3380 #+END_EXAMPLE
3382 You may also include an optional ``scale'' metadata number to
3383 specifiy the length of the touch feelers. The default is $0.1$,
3384 and this is generally sufficient.
3386 The touch UV should contain white pixels for each touch sensor.
3388 Here is an example touch-uv map that approximates a human finger,
3389 and its corresponding model.
3391 #+caption: This is the tactile-sensor-profile for the upper segment
3392 #+caption: of a fingertip. It defines regions of high touch sensitivity
3393 #+caption: (where there are many white pixels) and regions of low
3394 #+caption: sensitivity (where white pixels are sparse).
3395 #+name: guide-fingertip-UV
3396 #+ATTR_LaTeX: :width 9cm :placement [H]
3397 [[./images/finger-UV.png]]
3399 #+caption: The fingertip UV-image form above applied to a simple
3400 #+caption: model of a fingertip.
3401 #+name: guide-fingertip
3402 #+ATTR_LaTeX: :width 9cm :placement [H]
3403 [[./images/finger-2.png]]
3405 *** Propriocepotion
3407 Proprioception is tied to each joint node -- nothing special must
3408 be done in a blender model to enable proprioception other than
3409 creating joint nodes.
3411 *** Muscles
3413 Muscles are created by creating an empty node named =muscles= and
3414 then creating any number of empty child nodes to represent your
3415 creature's muscles.
3418 Muscle nodes should have the following metadata under the
3419 ``muscle'' label:
3421 #+BEGIN_EXAMPLE
3422 <muscle-profile-file-name>
3423 #+END_EXAMPLE
3425 Muscles should also have a ``strength'' metadata entry describing
3426 the muscle's total strength at full activation.
3428 Muscle profiles are simple images that contain the relative amount
3429 of muscle power in each simulated alpha motor neuron. The width of
3430 the image is the total size of the motor pool, and the redness of
3431 each neuron is the relative power of that motor pool.
3433 While the profile image can have any dimensions, only the first
3434 line of pixels is used to define the muscle. Here is a sample
3435 muscle profile image that defines a human-like muscle.
3437 #+caption: A muscle profile image that describes the strengths
3438 #+caption: of each motor neuron in a muscle. White is weakest
3439 #+caption: and dark red is strongest. This particular pattern
3440 #+caption: has weaker motor neurons at the beginning, just
3441 #+caption: like human muscle.
3442 #+name: muscle-recruit
3443 #+ATTR_LaTeX: :width 7cm :placement [H]
3444 [[./images/basic-muscle.png]]
3446 Muscles twist the nearest physical object about the muscle node's
3447 Z-axis. I recommend using the ``Single Arrow'' display mode for
3448 muscles and using the right hand rule to determine which way the
3449 muscle will twist. To make a segment that can twist in multiple
3450 directions, create multiple, differently aligned muscles.
3452 ** =CORTEX= API
3454 These are the some functions exposed by =CORTEX= for creating
3455 worlds and simulating creatures. These are in addition to
3456 jMonkeyEngine3's extensive library, which is documented elsewhere.
3458 *** Simulation
3459 - =(world root-node key-map setup-fn update-fn)= :: create
3460 a simulation.
3461 - /root-node/ :: a =com.jme3.scene.Node= object which
3462 contains all of the objects that should be in the
3463 simulation.
3465 - /key-map/ :: a map from strings describing keys to
3466 functions that should be executed whenever that key is
3467 pressed. the functions should take a SimpleApplication
3468 object and a boolean value. The SimpleApplication is the
3469 current simulation that is running, and the boolean is true
3470 if the key is being pressed, and false if it is being
3471 released. As an example,
3472 #+BEGIN_SRC clojure
3473 {"key-j" (fn [game value] (if value (println "key j pressed")))}
3474 #+END_SRC
3475 is a valid key-map which will cause the simulation to print
3476 a message whenever the 'j' key on the keyboard is pressed.
3478 - /setup-fn/ :: a function that takes a =SimpleApplication=
3479 object. It is called once when initializing the simulation.
3480 Use it to create things like lights, change the gravity,
3481 initialize debug nodes, etc.
3483 - /update-fn/ :: this function takes a =SimpleApplication=
3484 object and a float and is called every frame of the
3485 simulation. The float tells how many seconds is has been
3486 since the last frame was rendered, according to whatever
3487 clock jme is currently using. The default is to use IsoTimer
3488 which will result in this value always being the same.
3490 - =(position-camera world position rotation)= :: set the position
3491 of the simulation's main camera.
3493 - =(enable-debug world)= :: turn on debug wireframes for each
3494 simulated object.
3496 - =(set-gravity world gravity)= :: set the gravity of a running
3497 simulation.
3499 - =(box length width height & {options})= :: create a box in the
3500 simulation. Options is a hash map specifying texture, mass,
3501 etc. Possible options are =:name=, =:color=, =:mass=,
3502 =:friction=, =:texture=, =:material=, =:position=,
3503 =:rotation=, =:shape=, and =:physical?=.
3505 - =(sphere radius & {options})= :: create a sphere in the simulation.
3506 Options are the same as in =box=.
3508 - =(load-blender-model file-name)= :: create a node structure
3509 representing that described in a blender file.
3511 - =(light-up-everything world)= :: distribute a standard compliment
3512 of lights throught the simulation. Should be adequate for most
3513 purposes.
3515 - =(node-seq node)= :: return a recursuve list of the node's
3516 children.
3518 - =(nodify name children)= :: construct a node given a node-name and
3519 desired children.
3521 - =(add-element world element)= :: add an object to a running world
3522 simulation.
3524 - =(set-accuracy world accuracy)= :: change the accuracy of the
3525 world's physics simulator.
3527 - =(asset-manager)= :: get an /AssetManager/, a jMonkeyEngine
3528 construct that is useful for loading textures and is required
3529 for smooth interaction with jMonkeyEngine library functions.
3531 - =(load-bullet)= :: unpack native libraries and initialize
3532 blender. This function is required before other world building
3533 functions are called.
3535 *** Creature Manipulation / Import
3537 - =(body! creature)= :: give the creature a physical body.
3539 - =(vision! creature)= :: give the creature a sense of vision.
3540 Returns a list of functions which will each, when called
3541 during a simulation, return the vision data for the channel of
3542 one of the eyes. The functions are ordered depending on the
3543 alphabetical order of the names of the eye nodes in the
3544 blender file. The data returned by the functions is a vector
3545 containing the eye's /topology/, a vector of coordinates, and
3546 the eye's /data/, a vector of RGB values filtered by the eye's
3547 sensitivity.
3549 - =(hearing! creature)= :: give the creature a sense of hearing.
3550 Returns a list of functions, one for each ear, that when
3551 called will return a frame's worth of hearing data for that
3552 ear. The functions are ordered depending on the alphabetical
3553 order of the names of the ear nodes in the blender file. The
3554 data returned by the functions is an array PCM encoded wav
3555 data.
3557 - =(touch! creature)= :: give the creature a sense of touch. Returns
3558 a single function that must be called with the /root node/ of
3559 the world, and which will return a vector of /touch-data/
3560 one entry for each touch sensitive component, each entry of
3561 which contains a /topology/ that specifies the distribution of
3562 touch sensors, and the /data/, which is a vector of
3563 =[activation, length]= pairs for each touch hair.
3565 - =(proprioception! creature)= :: give the creature the sense of
3566 proprioception. Returns a list of functions, one for each
3567 joint, that when called during a running simulation will
3568 report the =[headnig, pitch, roll]= of the joint.
3570 - =(movement! creature)= :: give the creature the power of movement.
3571 Creates a list of functions, one for each muscle, that when
3572 called with an integer, will set the recruitment of that
3573 muscle to that integer, and will report the current power
3574 being exerted by the muscle. Order of muscles is determined by
3575 the alphabetical sort order of the names of the muscle nodes.
3577 *** Visualization/Debug
3579 - =(view-vision)= :: create a function that when called with a list
3580 of visual data returned from the functions made by =vision!=,
3581 will display that visual data on the screen.
3583 - =(view-hearing)= :: same as =view-vision= but for hearing.
3585 - =(view-touch)= :: same as =view-vision= but for touch.
3587 - =(view-proprioception)= :: same as =view-vision= but for
3588 proprioception.
3590 - =(view-movement)= :: same as =view-vision= but for
3591 proprioception.
3593 - =(view anything)= :: =view= is a polymorphic function that allows
3594 you to inspect almost anything you could reasonably expect to
3595 be able to ``see'' in =CORTEX=.
3597 - =(text anything)= :: =text= is a polymorphic function that allows
3598 you to convert practically anything into a text string.
3600 - =(println-repl anything)= :: print messages to clojure's repl
3601 instead of the simulation's terminal window.
3603 - =(mega-import-jme3)= :: for experimenting at the REPL. This
3604 function will import all jMonkeyEngine3 classes for immediate
3605 use.
3607 - =(display-dialated-time world timer)= :: Shows the time as it is
3608 flowing in the simulation on a HUD display.