Robot Learning from Human-Generated Rewards



To serve us well, robots and other agents must understand our needs and how to fulfill them. To that end, our research develops robots that empower humans by interactively learning from them. Interactive learning methods enable technically unskilled end-users to designate correct behavior and communicate their task knowledge to improve a robot's task performance. This research on interactive learning focuses on algorithms that facilitate teaching by signals of approval and disapproval from a live human trainer. We operationalize these feedback signals as numeric rewards within the machine-learning framework of reinforcement learning. In comparison to the complementary form of teaching by demonstration, this feedback-based teaching may require less task expertise and place less cognitive load on the trainer. Envisioned applications include human-robot collaboration and assistive robotic devices for handicapped users, such as myolectrically controlled prosthetics.