|
rlm@572
|
1 Here I demonstrate the power of using embodied artificial intelligence
|
|
rlm@572
|
2 to attack the \emph{action recognition} problem, which is the challenge of
|
|
rlm@572
|
3 recognizing actions performed by a creature given limited data about
|
|
rlm@572
|
4 the creature's actions, such as a video recording. I solve this
|
|
rlm@572
|
5 problem in the case of a worm-like creature performing actions such as
|
|
rlm@572
|
6 curling and wiggling.
|
|
rlm@572
|
7
|
|
rlm@572
|
8 To attack the action recognition problem, I developed a computational
|
|
rlm@572
|
9 model of empathy (\texttt{EMPATH}) which allows me to recognize actions using
|
|
rlm@572
|
10 simple, embodied representations of actions (which require rich
|
|
rlm@572
|
11 sensory data), even when that sensory data is not actually available.
|
|
rlm@572
|
12 The missing sense data is imagined by combining previous experiences
|
|
rlm@572
|
13 gained from unsupervised free play. The worm is a five-segment
|
|
rlm@572
|
14 creature equipped with touch, proprioception, and muscle tension
|
|
rlm@572
|
15 senses. It recognizes actions using only proprioception data.
|
|
rlm@572
|
16
|
|
rlm@572
|
17 In order to build this empathic, action-recognizing system, I created
|
|
rlm@572
|
18 a program called \texttt{CORTEX}, which is a complete platform for embodied
|
|
rlm@572
|
19 AI research. It provides multiple senses for simulated creatures,
|
|
rlm@572
|
20 including vision, touch, proprioception, muscle tension, and hearing.
|
|
rlm@572
|
21 Each of these senses provides a wealth of parameters that are
|
|
rlm@572
|
22 biologically inspired. \texttt{CORTEX} is able to simulate any number of
|
|
rlm@572
|
23 creatures and senses, and provides facilities for easily modeling and
|
|
rlm@572
|
24 creating new creatures. As a research platform it is more complete
|
|
rlm@572
|
25 than any other system currently available.
|
