Toward estimating others' transition models under occlusion for multi-robot IRL

摘要

Multi-robot inverse reinforcement learning (mIRL) is broadly useful for learning, from observations, the behaviors of multiple robots executing fixed tra-jectories and interacting with each other. In this pa-per, we relax a crucial assumption in IRL to make it better suited for wider robotic applications: we allow the transition functions of other robots to be stochastic and do not assume that the transi-tion error probabilities are known to the learner. Challenged by occlusion where large portions of others ’ state spaces are fully hidden, we present a new approach that maps stochastic transitions to distributions over features. Then, the undercon-strained problem is solved using nonlinear opti-mization that maximizes entropy to learn the tran-sition function of each robot from occluded obser-vations. Our methods represent significant and first steps toward making mIRL pragmatic. 1