eduzhai > Applied Sciences > Engineering >

To Share or Not to Share? Performance Guarantees and the Asymmetric Nature of Cross-Robot Experience Transfer

  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: In the robotics literature, experience transfer has been proposed indifferent learning-based control frameworks to minimize the costs and risksassociated with training robots. While various works have shown the feasibilityof transferring prior experience from a source robot to improve or acceleratethe learning of a target robot, there are usually no guarantees that experiencetransfer improves the performance of the target robot. In practice, theefficacy of transferring experience is often not known until it is tested onphysical robots. This trial-and-error approach can be extremely unsafe andinefficient. Building on our previous work, in this paper we consider aninverse module transfer learning framework, where the inverse module of asource robot system is transferred to a target robot system to improve itstracking performance on arbitrary trajectories. We derive a theoretical boundon the tracking error when a source inverse module is transferred to the targetrobot and propose a Bayesian-optimization-based algorithm to estimate thisbound from data. We further highlight the asymmetric nature of cross-robotexperience transfer that has often been neglected in the literature. Wedemonstrate our approach in quadrotor experiments and show that we canguarantee positive transfer on the target robot for tracking random periodictrajectories.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×