Generating a Symbolic Task Model from Multiple Demonstrations
Koichi Ogawara, Jun Takamatsu, Hiroshi Kimura, Katsushi Ikeuchi · 2002
Most of the approaches to ``Learning from Observation'' so far assume that a demonstration can be well understood from a single demonstration. But a single demonstration contains ambiguity, in that interactions which are essential to complete a task can't be discerned without prior task dependent knowledge, which should be obtained from observation. To address these issues, we propose a technique to integrate multiple observations of demonstrations and estimate essential interactions automatically. The demonstrations differ, but indicate virtually a same task. The shared interactions among all the demonstrations are considered to be essential and a task model is generated from them. Detected essential interactions are further clustered to form a set of primitive symbols and a symbolic task model is generated. Finally, the robot reproduces the same task based on the task model even in different environment.