We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Robotics

Title: LazyDAgger: Reducing Context Switching in Interactive Imitation Learning

Abstract: Corrective interventions while a robot is learning to automate a task provide an intuitive method for a human supervisor to assist the robot and convey information about desired behavior. However, these interventions can impose significant burden on a human supervisor, as each intervention interrupts other work the human is doing, incurs latency with each context switch between supervisor and autonomous control, and requires time to perform. We present LazyDAgger, which extends the interactive imitation learning (IL) algorithm SafeDAgger to reduce context switches between supervisor and autonomous control. We find that LazyDAgger improves the performance and robustness of the learned policy during both learning and execution while limiting burden on the supervisor. Simulation experiments suggest that LazyDAgger can reduce context switches by an average of 60% over SafeDAgger on 3 continuous control tasks while maintaining state-of-the-art policy performance. In physical fabric manipulation experiments with an ABB YuMi robot, LazyDAgger reduces context switches by 60% while achieving a 60% higher success rate than SafeDAgger at execution time.
Comments: IEEE CASE 2021
Subjects: Robotics (cs.RO); Artificial Intelligence (cs.AI)
Cite as: arXiv:2104.00053 [cs.RO]
  (or arXiv:2104.00053v2 [cs.RO] for this version)

Submission history

From: Ryan Hoque [view email]
[v1] Wed, 31 Mar 2021 18:22:53 GMT (21509kb,D)
[v2] Tue, 20 Jul 2021 21:47:14 GMT (43001kb,D)

Link back to: arXiv, form interface, contact.