Online incremental learning of manipulation tasks for semi-autonomous teleoperation

Conference item

We present an approach for online incremental learning of manipulation tasks. A Bayesian clustering algorithm is used to build an online hidden semi-Markov model (HSMM) that captures state transition and state duration probabilities of demonstrated motions. Motions are then generated by stochastically sampling from the learned model and tracked using an infinite-horizon linear quadratic regulator (LQR), with varying stiffness and damping characteristics learned from the demonstrations. Our approach provides a compact skill representation that is learned online and can be used in a semiautonomous teleoperation setting, where direct teleoperation is infeasible due to large communication latency or failure. We present a planar drawing example to highlight the flexibility of our approach, and demonstrate how our solution can be used for a hot-stabbing motion in an underwater teleoperation scenario. We evaluate the performance of our approach over multiple trials and report high accuracy and repeatability, resulting to consistently successful hot-stabbing motions.

Authors: Havoutis, I; Tanwani, A; Calinon, S

• Publication status: Not published
• Peer review status: Reviewed (other)
• Publisher: Institute of Electrical and Electronics Engineers
• Host title: International Conference on Intelligent Robots and Systems, Workshop on Closed Loop Grasping and Manipulation: Challenges and Progress
• Journal: International Conference on Intelligent Robots and Systems, Workshop on Closed Loop Grasping and Manipulation: Challenges and Progress
• Publication date: 2016-10-01
• Acceptance date: 2016-09-30