a real-time trajectory planning method for overhead crane based on deep reinforcement learning

Considering the trajectory planning requirements of overhead crane, a real-time trajectory planning method based on deep reinforcement learning is proposed. A trajectory planning model of overhead crane in transport phase is established, taking into account the random start positions of the overhead crane and the load uncertainty. The Twin delayed deep deterministic policy gradient algorithm (TD3) is used, in which the position and velocity of the planar crane and the sway angle and angular acceleration information between the crane and the load are used as the state quantities of the algorithm, and the thrust force applied on the planar crane is used as the control quantities, the proposed method has the generalization ability to meet the accuracy requirements, and the control method can make the crane in the process of movement with smooth speed, accurate positioning, small load sway angle, and has a certain degree of robustness in the face of the beginning of the position of the random and the load uncertainty, which can satisfy the needs of online real-time trajectory planning.