Computational elements of robot learning by imitation

Abstract

Robot learning by imitation makes an increasing body of robotics research. Imitation learning complements motor learning techniques by restricting the search space to a computationally tractable subset. Imitation learning search for spatial and temporal invariants across several demonstrations. These invariants are task- dependent. This talk will present an algorithm that determines the key features of an imitation task through a comparative analysis of the data in joint space, carthesian space and visual space. These features are used to control the reproduction of the observed motion by a 30 degrees of freedom humanoid robot.