Deep reinforcement learning control approach to mitigating actuator attacks.