What is imitation learning?
Imitation learning trains a policy from demonstrations rather than requiring every behavior to be specified manually. Demonstrations can encode task structure that is difficult to describe as rules.
Behavior cloning
Behavior cloning maps observations to demonstrated actions. It is simple and useful, but distribution shift can cause errors when the robot reaches states not represented in the demonstrations.
Why embodiment matters
The same demonstration may need different execution on different robots. Retargeting, shared representations and embodiment-aware policies can help bridge that gap.
Explore the field
Continue through the research map, model directory, robot directory and company directory.