Synthesizing Gatekeepers for Safe Reinforcement Learning

Demonstrates gatekeepers that block unsafe actions using model checking and neural control barrier functions to enable safe optimization.