Iterative Reachability Estimation for Safe Reinforcement Learning

We theoretically establish that our algorithms almost surely converge to locally optimal policies of our safe optimization framework.