Guiding Safe Exploration with Weakest Preconditions

Greg Anderson, Swarat Chaudhuri, Isil Dillig

In reinforcement learning for safety-critical settings, it is often desirable for the agent to obey safety constraints at all points in time, including during training. We present…