view thesis/abstract.org @ 539:fc116e960f56

more elaboration
author Robert McIntyre <rlm@mit.edu>
date Sun, 27 Apr 2014 21:52:39 -0400
parents 8e52a2802821
children d304b2ea7c58
line wrap: on
line source
1 Here I demonstrate the power of using embodied artificial intelligence
2 to attack the /action recognition/ problem, which is the challenge of
3 recognizing actions performed by a creature given limited data about
4 the creature's actions, such as a video recording. I solve this
5 problem in the case of a worm-like creature performing actions such as
6 curling and wiggling.
8 To attack the action recognition problem, I developed a computational
9 model of empathy (=EMPATH=) which allows me to recognize actions using
10 simple, embodied representations of actions (which require rich
11 sensory data), even when that sensory data is not actually available.
12 The missing sense data is ``imagined'' by the system by combining
13 previous experiences gained from unsupervised free play. The worm is a
14 five-segment creature equipped with touch, proprioception, and muscle
15 tension senses. It recognizes actions using only proprioception data.
17 In order to build this empathic, action-recognizing system, I created
18 a program called =CORTEX=, which is a complete platform for embodied
19 AI research. It provides multiple senses for simulated creatures,
20 including vision, touch, proprioception, muscle tension, and hearing.
21 Each of these senses provides a wealth of parameters that are
22 biologically inspired. =CORTEX= is able to simulate any number of
23 creatures and senses, and provides facilities for easily modeling and
24 creating new creatures. As a research platform it is more complete
25 than any other system currently available.