Learning Mobile Robot Motion Control from Demonstrated Primitives and Human Feedback

Abstract

Task demonstration is one effective technique for developing robot motion control policies. As tasks become more complex, however, demonstration can become more difficult. In this work we introduce a technique that uses corrective human feedback to build a policy able to perform an undemonstrated task from simpler policies learned from demonstration. Our algorithm first evaluates and corrects the execution of motion primitive policies learned from demonstration. The algorithm next corrects and enables the execution of a larger task built from these primitives. Within a simulated robot motion control domain, we validate that a policy for an undemonstrated task is successfully built from motion primitives learned from demonstration under our approach. We show feedback to both aid and enable policy development, improving policy performance in success, speed and efficiency.

Safe reinforcement learning in high-risk tasks through policy improvement

“…From this dataset, the learner then generalizes a control policy which maps observations to actions. Some works demonstrated the ability to improve a policy based on the experience from a teacher [4], [5]. In the case of the helicopter hover task, the competition software provided a baseline controller.…”