Direct Behavior Specification via Constrained Reinforcement Learning. (arXiv:2112.12228v2 [cs.LG] UPDATED)

The standard formulation of Reinforcement Learning lacks a practical way of
specifying what are admissible and forbidden behaviors. Most often,
practitioners go about the task of behavior specification by manually
engineering the reward function, a counter-intuitive process that requires
several iterations and is prone to reward hacking by the agent. In this work,
we argue that constrained RL, which has almost exclusively been used for safe
RL, also has the potential to significantly reduce the amount of work spent for …

arxiv learning reinforcement learning