Leveraging Implicit Human Feedback to Better Learn from Explicit Human Feedback in Human-Robot Interactions

摘要

My work aims to enable robots to more effectively learn how to help people. The way in which people want to be helped by robots can vary by task, person, or time, among other factors. Thus, it is important that robots can learn to tailor their behavior based on a person's evolving preferences during an interaction. Robots typically learn from humans via explicit feedback, such as evaluative feedback, preferences, demonstrations, or corrections. However, this type of feedback can interrupt the flow of an interaction and it places an additional cognitive burden on the human. We know that humans "leak" information through their non-verbal behavior that gives clues about their internal states during interactions? can this information be used to augment how a robot learns from humans? My research aims to explore how to incorporate feedback that humans provide implicitly into robot learning paradigms.