Multi-objective 3D Bin Packing strategies through Meta Reinforcement Learning

Efficient space utilization is a fundamental discipline in logistics, crucial for both storage and transportation of goods. The challenge of arranging goods geometrically to minimize empty space while maintaining stability is algorithmically represented by the three-dimensional bin packing problem. Existing literature predominantly focuses on volume minimization, neglecting other critical factors such as load stability, which is essential to prevent transport damage and associated costs. In this work, an approach based on meta reinforcement learning is presented, which enables the solution of multi-objective optimization problems in the 3D bin packing problem. The proposed solution and the training procedure are evaluated based on practical criteria, focusing on volume minimization and weight distribution. The experiments show that by using such methods, a flexible solution for the multi-objective 3D bin packing problem can be implemented.