A survey of inverse reinforcement learning: Challenges, methods and progress

Inverse reinforcement learning (IRL) is the problem of inferring reward function an agent, given its policy or observed behavior. Analogous to RL, IRL perceived both as a and class methods. By categorically surveying extant literature in IRL, this article serves comprehensive reference for researchers practitioners machine well those new it understand challenges select approaches best suited on hand. The survey formally introduces along with central such difficulty performing accurate inference generalizability, sensitivity prior knowledge, disproportionate growth solution complexity size. surveys vast collection foundational methods grouped together by commonality their objectives, elaborates how these mitigate challenges. We further discuss extensions traditional handling imperfect perception, incomplete model, multiple functions nonlinear functions. concludes discussion some broad advances research area currently open questions.