DDPG-Based PID Optimal Controller for Position and Sway Angle of RTGC

Increasing global trade requires fast and accurate container transfers by minimizing positional errors and swing angles on the Rubber Tyred Gantry Crane (RTGC) at the container yard causing that the optimization of the controller is a necessity. Currently developing artificial intelligence, namely reinforcement learning based Deep Deterministic Policy Gradient (DDPG) approach, can provide continuous action or output that can be used for controller optimization problems. DDPG can learn from its interactions with the environment, allowing it to identify the best optimal parameter region. This paper proposes DDPG as an PID controller optimizer for position and swing angle of RTGC tested on six episode variations and a trajectory reference. The results garnered from this proposed methodology reveal that the optimal PID parameters perform well across a spectrum of reference trajectories, as well as under varying system parameters such as container masses and rope lengths, as demonstrated by metrics like position and swing angle error, rise time, settling time, and overshoot.