Low-Complexity Physics-Informed Reinforcement Learning Using Post-Decision States With Stochastic Sampling