A Model-Based Reinforcement Learning Approach for Robust PID Tuning