A Deep Reinforcement Learning Based Optimization Scheme for Three-dimensional Lumber Crating

Currently, the loading and unloading process of lumber in the dock mostly relies on adding cables artificially to realize the function similar to that of a "container", but it often suffers from the problems of poor integrality, insufficient science, and weak safety. To address these problems, this scheme builds a three-dimensional grid environment to simulate the spatial structure of a container, and combines the Q-network structure based on the graph attention mechanism and the greedy algorithm to optimize the placement strategy of the lumber based on the real-time lumber length, diameter, and container state. This research provides new solutions for containerized cargo loading and new ideas for future terminal crating processes.