Optimizing 3D Bin Packing of Heterogeneous Objects Using Continuous Transformations in SE(3)

With the growth of online e-commerce platforms, the challenge of automatically packing objects into confined spaces has attracted increasing attention from the scientific community. This paper presents an algorithm tackling the bin packing problem of heterogeneous objects or, more precisely, the task of finding appropriate poses for items whose geometry can be traced back to primitive convex shapes, when these are to be inserted inside a box-shaped bin. Since this problem is strongly NP-complete, finding a solution in practical timeframes is not trivial for industrial applications, in which boxes must be filled in the span of seconds or minutes. This paper presents a heuristic-driven optimization problem that leverages a point-cloud representation of the bin and signed-distance functions of the items to be packed. Solution is sought in a continuous subset of , including both continuous translations and continuous rotations. To enhance robustness, the static stability of the items in the box is ensured through a mesh-based physics simulator. The proposed approach can be used, with suitable variants, for both offline and online packing. Performance is evaluated through simulations conducted within the physics simulator, evaluating the algorithm performance in different scenarios.