Autonomous Assessment of Demonstration Sufficiency via Bayesian Inverse Reinforcement Learning

Paper titled Autonomous Assessment of Demonstration Sufficiency via Bayesian Inverse Reinforcement Learning was selected for the upcoming 19th Annual ACM/IEEE International Conference on Human Robot Interaction (HRI 2024) that will be held from March 11-15, 2024 in Boulder, Colorado, USA.

In their paper, revised on 1/2/2024, the authors Tu Trinh, Haoyu Chen, and Daniel S. Brown evaluate their approach in simulation for both discrete and continuous state-space domains and illustrate the feasibility of developing a robotic system that can accurately evaluate demonstration sufficiency.

Abstract:

The authors of this paper examine the problem of determining demonstration sufficiency: how can a robot self-assess whether it has received enough demonstrations from an expert to ensure a desired level of performance? To address this, they propose a novel self-assessment approach based on Bayesian inverse reinforcement learning and high-confidence value-at-risk bounds on two definitions of sufficiency: (1) normalized expected value difference and (2) percent improvement over a baseline policy. Through simulation and user study, the authors show that using their approach a robot can accurately evaluate demonstration sufficiency and perform exactly as intended, without needing too many or perfectly optimal demonstrations, helping reduce human burden in such collaborative learning settings.