Evaluating a DDPG reinforcement learning agent on a ball-and-plate system: A comparative study of intelligent control approaches

This research investigates the performance of a novel deep reinforcement learning (DRL) agent in comparison with traditional intelligent control approaches for manipulating the tilt angles of ball-and-plate system. This study highlights the strengths and weaknesses of each technique through extensive experimentation and analysis of step response and trajectory tracking metrics while trailing a circular path. While the DRL agent demonstrates rapid responsiveness, it exhibits inferior trajectory tracking accuracy compared to the other methods, namely, Proportional-Integral-Derivative (PID), Model Predictive Control (MPC), Sliding Mode Control (SMC), and Linear Quadratic Regulator (LQR). These findings emphasize the importance of balancing speed and precision in control system design. Traditional methods like PID, MPC, and SMC showcase robust performance in achieving precise trajectory tracking with minimal error and overshoot, underscoring their suitability for practical applications. This comparative analysis contributes valuable insights for researchers and practitioners in control engineering, guiding the development of suitable control strategies for dynamic systems. Future research can consider hybrid control strategies that combine the strengths of traditional methods with reinforcement learning to achieve optimal tuning of the reinforcement learning agent for superior performance.