eduzhai > Applied Sciences > Engineering >

Learning from Sparse Demonstrations

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 16 pages

Abstract: This paper proposes an approach which enables a robot to learn an objectivefunction from sparse demonstrations of an expert. The demonstrations are givenby a small number of sparse waypoints; the waypoints are desired outputs of therobot s trajectory at certain time instances, sparsely located within ademonstration time horizon. The duration of the expert s demonstration may bedifferent from the actual duration of the robot s execution. The proposedmethod enables to jointly learn an objective function and a time-warpingfunction such that the robot s reproduced trajectory has minimal distance tothe sparse demonstration waypoints. Unlike existing inverse reinforcementlearning techniques, the proposed approach uses the differential Pontryagin smaximum principle, which allows direct minimization of the distance between therobot s trajectory and the sparse demonstration waypoints and enablessimultaneous learning of an objective function and a time-warping function. Wedemonstrate the effectiveness of the proposed approach in various simulatedscenarios. We apply the method to learn motion planning control of a 6-DoFmaneuvering unmanned aerial vehicle (UAV) and a robot arm in environments withobstacles. The results show that a robot is able to learn a valid objectivefunction to avoid obstacles with few demonstrated waypoints.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...