Reinforcement Learning with Adaptive Control Regularization for Safe Control of Critical Systems | Synapse