Mercurial > cortex

annotate thesis/org/roadmap.org @ 413:54ef2e06c3ef

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
gave main worm a consistent muscle order.
author Robert McIntyre <rlm@mit.edu>
date Wed, 19 Mar 2014 14:24:13 -0400
parents ea524e4d8f8d
children af7945c27474
rev   line source
rlm@401 1 In order for this to be a reasonable thesis that I can be proud of,
rlm@401 2 what are the /minimum/ number of things I need to get done?
rlm@401 3
rlm@401 4
rlm@401 5 * worm OR hand registration
rlm@401 6 - training from a few examples (2 to start out)
rlm@401 7 - aligning the body with the scene
rlm@401 8 - generating sensory data
rlm@401 9 - matching previous labeled examples using dot-products or some
rlm@401 10 other basic thing
rlm@401 11 - showing that it works with different views
rlm@401 12
rlm@401 13 * first draft
rlm@401 14 - draft of thesis without bibliography or formatting
rlm@401 15 - should have basic experiment and have full description of
rlm@401 16 framework with code
rlm@401 17 - review with Winston
rlm@401 18
rlm@401 19 * final draft
rlm@401 20 - implement stretch goals from Winston if possible
rlm@401 21 - complete final formatting and submit
rlm@401 22
rlm@401 23 * CORTEX
rlm@401 24 DEADLINE: <2014-05-09 Fri>
rlm@401 25 SHIT THAT'S IN 67 DAYS!!!
rlm@401 26
rlm@403 27 ** program simple feature matching code for the worm's segments
rlm@403 28
rlm@401 29 Subgoals:
rlm@401 30 *** DONE Get cortex working again, run tests, no jmonkeyengine updates
rlm@401 31 CLOSED: [2014-03-03 Mon 22:07] SCHEDULED: <2014-03-03 Mon>
rlm@401 32 *** DONE get blender working again
rlm@401 33 CLOSED: [2014-03-03 Mon 22:43] SCHEDULED: <2014-03-03 Mon>
rlm@401 34 *** DONE make sparce touch worm segment in blender
rlm@401 35 CLOSED: [2014-03-03 Mon 23:16] SCHEDULED: <2014-03-03 Mon>
rlm@401 36 CLOCK: [2014-03-03 Mon 22:44]--[2014-03-03 Mon 23:16] => 0:32
rlm@401 37 *** DONE make multi-segment touch worm with touch sensors and display
rlm@401 38 CLOSED: [2014-03-03 Mon 23:54] SCHEDULED: <2014-03-03 Mon>
rlm@401 39
rlm@401 40 *** DONE Make a worm wiggle and curl
rlm@401 41 CLOSED: [2014-03-04 Tue 23:03] SCHEDULED: <2014-03-04 Tue>
rlm@403 42
rlm@401 43
rlm@401 44 ** First draft
rlm@403 45
rlm@401 46 Subgoals:
rlm@401 47 *** Writeup new worm experiments.
rlm@401 48 *** Triage implementation code and get it into chapter form.
rlm@401 49
rlm@401 50
rlm@401 51
rlm@401 52
rlm@401 53
rlm@401 54 ** for today
rlm@401 55
rlm@401 56 - guided worm :: control the worm with the keyboard. Useful for
rlm@401 57 testing the body-centered recog scripts, and for
rlm@401 58 preparing a cool demo video.
rlm@401 59
rlm@401 60 - body-centered recognition :: detect actions using hard coded
rlm@401 61 body-centered scripts.
rlm@401 62
rlm@401 63 - cool demo video of the worm being moved and recognizing things ::
rlm@401 64 will be a neat part of the thesis.
rlm@401 65
rlm@401 66 - thesis export :: refactoring and organization of code so that it
rlm@401 67 spits out a thesis in addition to the web page.
rlm@401 68
rlm@401 69 - video alignment :: analyze the frames of a video in order to align
rlm@401 70 the worm. Requires body-centered recognition. Can "cheat".
rlm@401 71
rlm@401 72 - smoother actions :: use debugging controls to directly influence the
rlm@401 73 demo actions, and to generate recoginition procedures.
rlm@401 74
rlm@401 75 - degenerate video demonstration :: show the system recognizing a
rlm@401 76 curled worm from dead on. Crowning achievement of thesis.
rlm@401 77
rlm@401 78 ** Ordered from easiest to hardest
rlm@401 79
rlm@401 80 Just report the positions of everything. I don't think that this
rlm@401 81 necessairly shows anything usefull.
rlm@401 82
rlm@401 83 Worm-segment vision -- you initialize a view of the worm, but instead
rlm@401 84 of pixels you use labels via ray tracing. Has the advantage of still
rlm@401 85 allowing for visual occlusion, but reliably identifies the objects,
rlm@401 86 even without rainbow coloring. You can code this as an image.
rlm@401 87
rlm@401 88 Same as above, except just with worm/non-worm labels.
rlm@401 89
rlm@401 90 Color code each worm segment and then recognize them using blob
rlm@401 91 detectors. Then you solve for the perspective and the action
rlm@401 92 simultaneously.
rlm@401 93
rlm@401 94 The entire worm can be colored the same, high contrast color against a
rlm@401 95 nearly black background.
rlm@401 96
rlm@401 97 "Rooted" vision. You give the exact coordinates of ONE piece of the
rlm@401 98 worm, but the algorithm figures out the rest.
rlm@401 99
rlm@401 100 More rooted vision -- start off the entire worm with one posistion.
rlm@401 101
rlm@401 102 The right way to do alignment is to use motion over multiple frames to
rlm@401 103 snap individual pieces of the model into place sharing and
rlm@401 104 propragating the individual alignments over the whole model. We also
rlm@401 105 want to limit the alignment search to just those actions we are
rlm@401 106 prepared to identify. This might mean that I need some small "micro
rlm@401 107 actions" such as the individual movements of the worm pieces.
rlm@401 108
rlm@401 109 Get just the centers of each segment projected onto the imaging
rlm@401 110 plane. (best so far).
rlm@401 111
rlm@401 112
rlm@401 113 Repertoire of actions + video frames -->
rlm@401 114 directed multi-frame-search alg
rlm@401 115
rlm@401 116
rlm@401 117
rlm@401 118
rlm@401 119
rlm@401 120
rlm@401 121 !! Could also have a bounding box around the worm provided by
rlm@401 122 filtering the worm/non-worm render, and use bbbgs. As a bonus, I get
rlm@401 123 to include bbbgs in my thesis! Could finally do that recursive things
rlm@401 124 where I make bounding boxes be those things that give results that
rlm@401 125 give good bounding boxes. If I did this I could use a disruptive
rlm@401 126 pattern on the worm.
rlm@401 127
rlm@401 128 Re imagining using default textures is very simple for this system,
rlm@401 129 but hard for others.
rlm@401 130
rlm@401 131
rlm@401 132 Want to demonstrate, at minimum, alignment of some model of the worm
rlm@401 133 to the video, and a lookup of the action by simulated perception.
rlm@401 134
rlm@401 135 note: the purple/white points is a very beautiful texture, because
rlm@401 136 when it moves slightly, the white dots look like they're
rlm@401 137 twinkling. Would look even better if it was a darker purple. Also
rlm@401 138 would look better more spread out.
rlm@401 139
rlm@401 140
rlm@401 141 embed assumption of one frame of view, search by moving around in
rlm@401 142 simulated world.
rlm@401 143
rlm@401 144 Allowed to limit search by setting limits to a hemisphere around the
rlm@401 145 imagined worm! This limits scale also.
rlm@401 146
rlm@401 147
rlm@401 148
rlm@401 149
rlm@401 150
rlm@401 151 !! Limited search with worm/non-worm rendering.
rlm@401 152 How much inverse kinematics do we have to do?
rlm@401 153 What about cached (allowed state-space) paths, derived from labeled
rlm@401 154 training. You have to lead from one to another.
rlm@401 155
rlm@401 156 What about initial state? Could start the input videos at a specific
rlm@401 157 state, then just match that explicitly.
rlm@401 158
rlm@401 159 !! The training doesn't have to be labeled -- you can just move around
rlm@401 160 for a while!!
rlm@401 161
rlm@401 162 !! Limited search with motion based alignment.
rlm@401 163
rlm@401 164
rlm@401 165
rlm@401 166
rlm@401 167 "play arounds" can establish a chain of linked sensoriums. Future
rlm@401 168 matches must fall into one of the already experienced things, and once
rlm@401 169 they do, it greatly limits the things that are possible in the future.
rlm@401 170
rlm@401 171
rlm@401 172 frame differences help to detect muscle exertion.
rlm@401 173
rlm@401 174 Can try to match on a few "representative" frames. Can also just have
rlm@401 175 a few "bodies" in various states which we try to match.
rlm@401 176
rlm@401 177
rlm@401 178
rlm@401 179 Paths through state-space have the exact same signature as
rlm@401 180 simulation. BUT, these can be searched in parallel and don't interfere
rlm@401 181 with each other.
rlm@401 182
rlm@401 183
rlm@402 184
rlm@402 185
rlm@402 186 ** Final stretch up to First Draft
rlm@402 187
rlm@404 188 *** DONE complete debug control of worm
rlm@404 189 CLOSED: [2014-03-17 Mon 17:29] SCHEDULED: <2014-03-17 Mon>
rlm@404 190 CLOCK: [2014-03-17 Mon 14:01]--[2014-03-17 Mon 17:29] => 3:28
rlm@405 191 *** DONE add phi-space output to debug control
rlm@405 192 CLOSED: [2014-03-17 Mon 17:42] SCHEDULED: <2014-03-17 Mon>
rlm@405 193 CLOCK: [2014-03-17 Mon 17:31]--[2014-03-17 Mon 17:42] => 0:11
rlm@407 194
rlm@409 195 *** DONE complete automatic touch partitioning
rlm@409 196 CLOSED: [2014-03-18 Tue 21:43] SCHEDULED: <2014-03-18 Tue>
rlm@407 197 *** TODO complete cyclic predicate
rlm@409 198 SCHEDULED: <2014-03-18 Tue>
rlm@405 199 *** TODO complete three phi-stream action predicatates; test them with debug control
rlm@403 200 SCHEDULED: <2014-03-17 Mon>
rlm@407 201 CLOCK: [2014-03-18 Tue 18:36]
rlm@407 202 CLOCK: [2014-03-18 Tue 18:34]--[2014-03-18 Tue 18:36] => 0:02
rlm@407 203 CLOCK: [2014-03-17 Mon 19:19]--[2014-03-17 Mon 21:19] => 2:00
rlm@403 204 *** TODO create test videos, also record positions of worm segments
rlm@403 205 SCHEDULED: <2014-03-17 Mon>
rlm@403 206 *** TODO complete proprioception based movement lookup in phi-space
rlm@403 207 SCHEDULED: <2014-03-17 Mon>
rlm@402 208
rlm@402 209
rlm@403 210 *** TODO Collect intro, worm-learn and cortex creation into draft thesis.
rlm@405 211