Anti-swing overhead crane control algorithm based on multi-body dynamics using reinforcement learning

Abstract

Excessive cargo sway during crane operations in the current shipbuilding industry is a major problem that causes safety accidents and work delays. Therefore, the development of stable crane control technology is essential. In this study, a crane control algorithm that simultaneously achieves accurate movement to target positions and sway minimization was developed using reinforcement learning. In dynamics modeling, the Discrete Euler-Lagrange Equation was applied to significantly reduce the computational complexity of existing methods, and the Proximal Policy Optimization (PPO) method was used for control policy learning. A three-dimensional virtual environment was constructed to perform learning under various travel distances and operating conditions, and the performance of the developed algorithm was compared and verified against the traditional trapezoidal velocity profile. Experimental results showed that the proposed method exhibited significant improvements in position control precision and sway suppression performance compared to existing methods. The results of this study are expected to contribute to the implementation of automated crane control systems in actual shipyard environments.