view org/sense.org @ 230:f9b7d674aed8

reorganizing touch
author Robert McIntyre <rlm@mit.edu>
date Sat, 11 Feb 2012 19:28:36 -0700
parents 7c374c6cfe17
children be78d7bd6920
line wrap: on
line source
1 #+title: Helper Functions / Motivations
2 #+author: Robert McIntyre
3 #+email: rlm@mit.edu
4 #+description: sensory utilities
5 #+keywords: simulation, jMonkeyEngine3, clojure, simulated senses
6 #+SETUPFILE: ../../aurellem/org/setup.org
7 #+INCLUDE: ../../aurellem/org/level-0.org
10 * Blender Utilities
11 In blender, any object can be assigned an arbitray number of key-value
12 pairs which are called "Custom Properties". These are accessable in
13 jMonkyeEngine when blender files are imported with the
14 =BlenderLoader=. =(meta-data)= extracts these properties.
16 #+name: blender-1
17 #+begin_src clojure
18 (defn meta-data
19 "Get the meta-data for a node created with blender."
20 [blender-node key]
21 (if-let [data (.getUserData blender-node "properties")]
22 (.findValue data key) nil))
23 #+end_src
25 Blender uses a different coordinate system than jMonkeyEngine so it
26 is useful to be able to convert between the two. These only come into
27 play when the meta-data of a node refers to a vector in the blender
28 coordinate system.
30 #+name: blender-2
31 #+begin_src clojure
32 (defn jme-to-blender
33 "Convert from JME coordinates to Blender coordinates"
34 [#^Vector3f in]
35 (Vector3f. (.getX in) (- (.getZ in)) (.getY in)))
37 (defn blender-to-jme
38 "Convert from Blender coordinates to JME coordinates"
39 [#^Vector3f in]
40 (Vector3f. (.getX in) (.getZ in) (- (.getY in))))
41 #+end_src
43 * Sense Topology
45 Human beings are three-dimensional objects, and the nerves that
46 transmit data from our various sense organs to our brain are
47 essentially one-dimensional. This leaves up to two dimensions in which
48 our sensory information may flow. For example, imagine your skin: it
49 is a two-dimensional surface around a three-dimensional object (your
50 body). It has discrete touch sensors embedded at various points, and
51 the density of these sensors corresponds to the sensitivity of that
52 region of skin. Each touch sensor connects to a nerve, all of which
53 eventually are bundled together as they travel up the spinal cord to
54 the brain. Intersect the spinal nerves with a guillotining plane and
55 you will see all of the sensory data of the skin revealed in a roughly
56 circular two-dimensional image which is the cross section of the
57 spinal cord. Points on this image that are close together in this
58 circle represent touch sensors that are /probably/ close together on
59 the skin, although there is of course some cutting and rerangement
60 that has to be done to transfer the complicated surface of the skin
61 onto a two dimensional image.
63 Most human senses consist of many discrete sensors of various
64 properties distributed along a surface at various densities. For
65 skin, it is Pacinian corpuscles, Meissner's corpuscles, Merkel's
66 disks, and Ruffini's endings, which detect pressure and vibration of
67 various intensities. For ears, it is the stereocilia distributed
68 along the basilar membrane inside the cochlea; each one is sensitive
69 to a slightly different frequency of sound. For eyes, it is rods
70 and cones distributed along the surface of the retina. In each case,
71 we can describe the sense with a surface and a distribution of sensors
72 along that surface.
74 ** UV-maps
76 Blender and jMonkeyEngine already have support for exactly this sort
77 of data structure because it is used to "skin" models for games. It is
78 called [[http://wiki.blender.org/index.php/Doc:2.6/Manual/Textures/Mapping/UV][UV-mapping]]. The three-dimensional surface of a model is cut
79 and smooshed until it fits on a two-dimensional image. You paint
80 whatever you want on that image, and when the three-dimensional shape
81 is rendered in a game the smooshing and cutting us reversed and the
82 image appears on the three-dimensional object.
84 To make a sense, interpret the UV-image as describing the distribution
85 of that senses sensors. To get different types of sensors, you can
86 either use a different color for each type of sensor, or use multiple
87 UV-maps, each labeled with that sensor type. I generally use a white
88 pixel to mean the presense of a sensor and a black pixel to mean the
89 absense of a sensor, and use one UV-map for each sensor-type within a
90 given sense. The paths to the images are not stored as the actual
91 UV-map of the blender object but are instead referenced in the
92 meta-data of the node.
94 #+CAPTION: The UV-map for an enlongated icososphere. The white dots each represent a touch sensor. They are dense in the regions that describe the tip of the finger, and less dense along the dorsal side of the finger opposite the tip.
95 #+ATTR_HTML: width="300"
96 [[../images/finger-UV.png]]
98 #+CAPTION: Ventral side of the UV-mapped finger. Notice the density of touch sensors at the tip.
99 #+ATTR_HTML: width="300"
100 [[../images/finger-1.png]]
102 #+CAPTION: Side view of the UV-mapped finger.
103 #+ATTR_HTML: width="300"
104 [[../images/finger-2.png]]
106 #+CAPTION: Head on view of the finger. In both the head and side views you can see the divide where the touch-sensors transition from high density to low density.
107 #+ATTR_HTML: width="300"
108 [[../images/finger-3.png]]
110 The following code loads images and gets the locations of the white
111 pixels so that they can be used to create senses. =(load-image)= finds
112 images using jMonkeyEngine's asset-manager, so the image path is
113 expected to be relative to the =assets= directory. Thanks to Dylan
114 for the beautiful version of =(filter-pixels)=.
116 #+name: topology-1
117 #+begin_src clojure
118 (defn load-image
119 "Load an image as a BufferedImage using the asset-manager system."
120 [asset-relative-path]
121 (ImageToAwt/convert
122 (.getImage (.loadTexture (asset-manager) asset-relative-path))
123 false false 0))
125 (def white 0xFFFFFF)
127 (defn white? [rgb]
128 (= (bit-and white rgb) white))
130 (defn filter-pixels
131 "List the coordinates of all pixels matching pred, within the bounds
132 provided. If bounds are not specified then the entire image is
133 searched.
134 bounds -> [x0 y0 width height]"
135 {:author "Dylan Holmes"}
136 ([pred #^BufferedImage image]
137 (filter-pixels pred image [0 0 (.getWidth image) (.getHeight image)]))
138 ([pred #^BufferedImage image [x0 y0 width height]]
139 ((fn accumulate [x y matches]
140 (cond
141 (>= y (+ height y0)) matches
142 (>= x (+ width x0)) (recur 0 (inc y) matches)
143 (pred (.getRGB image x y))
144 (recur (inc x) y (conj matches [x y]))
145 :else (recur (inc x) y matches)))
146 x0 y0 [])))
148 (defn white-coordinates
149 "Coordinates of all the white pixels in a subset of the image."
150 ([#^BufferedImage image bounds]
151 (filter-pixels white? image bounds))
152 ([#^BufferedImage image]
153 (filter-pixels white? image)))
154 #+end_src
156 ** Topology
158 Information from the senses is transmitted to the brain via bundles of
159 axons, whether it be the optic nerve or the spinal cord. While these
160 bundles more or less perserve the overall topology of a sense's
161 two-dimensional surface, they do not perserve the percise euclidean
162 distances between every sensor. =(collapse)= is here to smoosh the
163 sensors described by a UV-map into a contigous region that still
164 perserves the topology of the original sense.
166 #+name: topology-2
167 #+begin_src clojure
168 (defn average [coll]
169 (/ (reduce + coll) (count coll)))
171 (defn collapse-1d
172 "One dimensional analogue of collapse."
173 [center line]
174 (let [length (count line)
175 num-above (count (filter (partial < center) line))
176 num-below (- length num-above)]
177 (range (- center num-below)
178 (+ center num-above))))
180 (defn collapse
181 "Take a set of pairs of integers and collapse them into a
182 contigous bitmap with no \"holes\"."
183 [points]
184 (if (empty? points) []
185 (let
186 [num-points (count points)
187 center (vector
188 (int (average (map first points)))
189 (int (average (map first points))))
190 flattened
191 (reduce
192 concat
193 (map
194 (fn [column]
195 (map vector
196 (map first column)
197 (collapse-1d (second center)
198 (map second column))))
199 (partition-by first (sort-by first points))))
200 squeezed
201 (reduce
202 concat
203 (map
204 (fn [row]
205 (map vector
206 (collapse-1d (first center)
207 (map first row))
208 (map second row)))
209 (partition-by second (sort-by second flattened))))
210 relocated
211 (let [min-x (apply min (map first squeezed))
212 min-y (apply min (map second squeezed))]
213 (map (fn [[x y]]
214 [(- x min-x)
215 (- y min-y)])
216 squeezed))]
217 relocated)))
218 #+end_src
219 * Viewing Sense Data
221 It's vital to /see/ the sense data to make sure that everything is
222 behaving as it should. =(view-sense)= and its helper, =(view-image)=
223 are here so that each sense can define its own way of turning
224 sense-data into pictures, while the actual rendering of said pictures
225 stays in one central place. =(points->image)= helps senses generate a
226 base image onto which they can overlay actual sense data.
228 #+name: view-senses
229 #+begin_src clojure
230 (in-ns 'cortex.sense)
232 (defn view-image
233 "Initailizes a JPanel on which you may draw a BufferedImage.
234 Returns a function that accepts a BufferedImage and draws it to the
235 JPanel. If given a directory it will save the images as png files
236 starting at 0000000.png and incrementing from there."
237 ([#^File save]
238 (let [idx (atom -1)
239 image
240 (atom
241 (BufferedImage. 1 1 BufferedImage/TYPE_4BYTE_ABGR))
242 panel
243 (proxy [JPanel] []
244 (paint
245 [graphics]
246 (proxy-super paintComponent graphics)
247 (.drawImage graphics @image 0 0 nil)))
248 frame (JFrame. "Display Image")]
249 (SwingUtilities/invokeLater
250 (fn []
251 (doto frame
252 (-> (.getContentPane) (.add panel))
253 (.pack)
254 (.setLocationRelativeTo nil)
255 (.setResizable true)
256 (.setVisible true))))
257 (fn [#^BufferedImage i]
258 (reset! image i)
259 (.setSize frame (+ 8 (.getWidth i)) (+ 28 (.getHeight i)))
260 (.repaint panel 0 0 (.getWidth i) (.getHeight i))
261 (if save
262 (ImageIO/write
263 i "png"
264 (File. save (format "%07d.png" (swap! idx inc))))))))
265 ([] (view-image nil)))
267 (defn view-sense
268 "Take a kernel that produces a BufferedImage from some sense data
269 and return a function which takes a list of sense data, uses the
270 kernel to convert to images, and displays those images, each in
271 its own JFrame."
272 [sense-display-kernel]
273 (let [windows (atom [])]
274 (fn this
275 ([data]
276 (this data nil))
277 ([data save-to]
278 (if (> (count data) (count @windows))
279 (reset!
280 windows
281 (doall
282 (map
283 (fn [idx]
284 (if save-to
285 (let [dir (File. save-to (str idx))]
286 (.mkdir dir)
287 (view-image dir))
288 (view-image))) (range (count data))))))
289 (dorun
290 (map
291 (fn [display datum]
292 (display (sense-display-kernel datum)))
293 @windows data))))))
296 (defn points->image
297 "Take a collection of points and visuliaze it as a BufferedImage."
298 [points]
299 (if (empty? points)
300 (BufferedImage. 1 1 BufferedImage/TYPE_BYTE_BINARY)
301 (let [xs (vec (map first points))
302 ys (vec (map second points))
303 x0 (apply min xs)
304 y0 (apply min ys)
305 width (- (apply max xs) x0)
306 height (- (apply max ys) y0)
307 image (BufferedImage. (inc width) (inc height)
308 BufferedImage/TYPE_INT_RGB)]
309 (dorun
310 (for [x (range (.getWidth image))
311 y (range (.getHeight image))]
312 (.setRGB image x y 0xFF0000)))
313 (dorun
314 (for [index (range (count points))]
315 (.setRGB image (- (xs index) x0) (- (ys index) y0) -1)))
316 image)))
318 (defn gray
319 "Create a gray RGB pixel with R, G, and B set to num. num must be
320 between 0 and 255."
321 [num]
322 (+ num
323 (bit-shift-left num 8)
324 (bit-shift-left num 16)))
325 #+end_src
327 * Building a Sense from Nodes
328 My method for defining senses in blender is the following:
330 Senses like vision and hearing are localized to a single point
331 and follow a particular object around. For these:
333 - Create a single top-level empty node whose name is the name of the sense
334 - Add empty nodes which each contain meta-data relevant
335 to the sense, including a UV-map describing the number/distribution
336 of sensors if applicipable.
337 - Make each empty-node the child of the top-level
338 node. =(sense-nodes)= below generates functions to find these children.
340 For touch, store the path to the UV-map which describes touch-sensors in the
341 meta-data of the object to which that map applies.
343 Each sense provides code that analyzes the Node structure of the
344 creature and creates sense-functions. They also modify the Node
345 structure if necessary.
347 Empty nodes created in blender have no appearance or physical presence
348 in jMonkeyEngine, but do appear in the scene graph. Empty nodes that
349 represent a sense which "follows" another geometry (like eyes and
350 ears) follow the closest physical object. =(closest-node)= finds this
351 closest object given the Creature and a particular empty node.
353 #+name: node-1
354 #+begin_src clojure
355 (defn sense-nodes
356 "For some senses there is a special empty blender node whose
357 children are considered markers for an instance of that sense. This
358 function generates functions to find those children, given the name
359 of the special parent node."
360 [parent-name]
361 (fn [#^Node creature]
362 (if-let [sense-node (.getChild creature parent-name)]
363 (seq (.getChildren sense-node))
364 (do (println-repl "could not find" parent-name "node") []))))
366 (defn closest-node
367 "Return the physical node in creature which is closest to the given
368 node."
369 [#^Node creature #^Node empty]
370 (loop [radius (float 0.01)]
371 (let [results (CollisionResults.)]
372 (.collideWith
373 creature
374 (BoundingBox. (.getWorldTranslation empty)
375 radius radius radius)
376 results)
377 (if-let [target (first results)]
378 (.getGeometry target)
379 (recur (float (* 2 radius)))))))
381 (defn world-to-local
382 "Convert the world coordinates into coordinates relative to the
383 object (i.e. local coordinates), taking into account the rotation
384 of object."
385 [#^Spatial object world-coordinate]
386 (.worldToLocal object world-coordinate nil))
388 (defn local-to-world
389 "Convert the local coordinates into world relative coordinates"
390 [#^Spatial object local-coordinate]
391 (.localToWorld object local-coordinate nil))
392 #+end_src
394 ** Sense Binding
396 =(bind-sense)= binds either a Camera or a Listener object to any
397 object so that they will follow that object no matter how it
398 moves. It is used to create both eyes and ears.
400 #+name: node-2
401 #+begin_src clojure
402 (defn bind-sense
403 "Bind the sense to the Spatial such that it will maintain its
404 current position relative to the Spatial no matter how the spatial
405 moves. 'sense can be either a Camera or Listener object."
406 [#^Spatial obj sense]
407 (let [sense-offset (.subtract (.getLocation sense)
408 (.getWorldTranslation obj))
409 initial-sense-rotation (Quaternion. (.getRotation sense))
410 base-anti-rotation (.inverse (.getWorldRotation obj))]
411 (.addControl
412 obj
413 (proxy [AbstractControl] []
414 (controlUpdate [tpf]
415 (let [total-rotation
416 (.mult base-anti-rotation (.getWorldRotation obj))]
417 (.setLocation
418 sense
419 (.add
420 (.mult total-rotation sense-offset)
421 (.getWorldTranslation obj)))
422 (.setRotation
423 sense
424 (.mult total-rotation initial-sense-rotation))))
425 (controlRender [_ _])))))
426 #+end_src
428 Here is some example code which shows how a camera bound to a blue box
429 with =(bind-sense)= moves as the box is buffeted by white cannonballs.
431 #+name: test
432 #+begin_src clojure
433 (defn test-bind-sense
434 "Show a camera that stays in the same relative position to a blue
435 cube."
436 []
437 (let [eye-pos (Vector3f. 0 30 0)
438 rock (box 1 1 1 :color ColorRGBA/Blue
439 :position (Vector3f. 0 10 0)
440 :mass 30)
441 table (box 3 1 10 :color ColorRGBA/Gray :mass 0
442 :position (Vector3f. 0 -3 0))]
443 (world
444 (nodify [rock table])
445 standard-debug-controls
446 (fn init [world]
447 (let [cam (doto (.clone (.getCamera world))
448 (.setLocation eye-pos)
449 (.lookAt Vector3f/ZERO
450 Vector3f/UNIT_X))]
451 (bind-sense rock cam)
452 (.setTimer world (RatchetTimer. 60))
453 (Capture/captureVideo
454 world (File. "/home/r/proj/cortex/render/bind-sense0"))
455 (add-camera!
456 world cam
457 (comp (view-image
458 (File. "/home/r/proj/cortex/render/bind-sense1"))
459 BufferedImage!))
460 (add-camera! world (.getCamera world) no-op)))
461 no-op)))
462 #+end_src
464 #+begin_html
465 <video controls="controls" width="755">
466 <source src="../video/bind-sense.ogg" type="video/ogg"
467 preload="none" poster="../images/aurellem-1280x480.png" />
468 </video>
469 #+end_html
471 With this, eyes are easy --- you just bind the camera closer to the
472 desired object, and set it to look outward instead of inward as it
473 does in the video.
475 (nb : the video was created with the following commands)
477 *** Combine Frames with ImageMagick
478 #+begin_src clojure :results silent
479 (ns cortex.video.magick
480 (:import java.io.File)
481 (:use clojure.contrib.shell-out))
483 (defn combine-images []
484 (let
485 [idx (atom -1)
486 left (rest
487 (sort
488 (file-seq (File. "/home/r/proj/cortex/render/bind-sense0/"))))
489 right (rest
490 (sort
491 (file-seq
492 (File. "/home/r/proj/cortex/render/bind-sense1/"))))
493 sub (rest
494 (sort
495 (file-seq
496 (File. "/home/r/proj/cortex/render/bind-senseB/"))))
497 sub* (concat sub (repeat 1000 (last sub)))]
498 (dorun
499 (map
500 (fn [im-1 im-2 sub]
501 (sh "convert" (.getCanonicalPath im-1)
502 (.getCanonicalPath im-2) "+append"
503 (.getCanonicalPath sub) "-append"
504 (.getCanonicalPath
505 (File. "/home/r/proj/cortex/render/bind-sense/"
506 (format "%07d.png" (swap! idx inc))))))
507 left right sub*))))
508 #+end_src
510 *** Encode Frames with ffmpeg
512 #+begin_src sh :results silent
513 cd /home/r/proj/cortex/render/
514 ffmpeg -r 30 -i bind-sense/%07d.png -b:v 9000k -vcodec libtheora bind-sense.ogg
515 #+end_src
517 * Headers
518 #+name: sense-header
519 #+begin_src clojure
520 (ns cortex.sense
521 "Here are functions useful in the construction of two or more
522 sensors/effectors."
523 {:author "Robert McInytre"}
524 (:use (cortex world util))
525 (:import ij.process.ImageProcessor)
526 (:import jme3tools.converters.ImageToAwt)
527 (:import java.awt.image.BufferedImage)
528 (:import com.jme3.collision.CollisionResults)
529 (:import com.jme3.bounding.BoundingBox)
530 (:import (com.jme3.scene Node Spatial))
531 (:import com.jme3.scene.control.AbstractControl)
532 (:import (com.jme3.math Quaternion Vector3f))
533 (:import javax.imageio.ImageIO)
534 (:import java.io.File)
535 (:import (javax.swing JPanel JFrame SwingUtilities)))
536 #+end_src
538 #+name: test-header
539 #+begin_src clojure
540 (ns cortex.test.sense
541 (:use (cortex world util sense vision))
542 (:import
543 java.io.File
544 (com.jme3.math Vector3f ColorRGBA)
545 (com.aurellem.capture RatchetTimer Capture)))
546 #+end_src
548 * Source Listing
549 - [[../src/cortex/sense.clj][cortex.sense]]
550 - [[../src/cortex/test/sense.clj][cortex.test.sense]]
551 - [[../assets/Models/subtitles/subtitles.blend][subtitles.blend]]
552 - [[../assets/Models/subtitles/Lake_CraterLake03_sm.hdr][subtitles reflection map]]
553 #+html: <ul> <li> <a href="../org/sense.org">This org file</a> </li> </ul>
554 - [[http://hg.bortreb.com ][source-repository]]
556 * Next
557 Now that some of the preliminaries are out of the way, in the [[./body.org][next
558 post]] I'll create a simulated body.
561 * COMMENT generate source
562 #+begin_src clojure :tangle ../src/cortex/sense.clj
563 <<sense-header>>
564 <<blender-1>>
565 <<blender-2>>
566 <<topology-1>>
567 <<topology-2>>
568 <<node-1>>
569 <<node-2>>
570 <<view-senses>>
571 #+end_src
573 #+begin_src clojure :tangle ../src/cortex/test/sense.clj
574 <<test-header>>
575 <<test>>
576 #+end_src
578 #+begin_src clojure :tangle ../src/cortex/video/magick.clj
579 <<magick>>
580 #+end_src