Differentiable Packing of Irregular 3D Objects with Adaptive Container Estimation

2026-06-15

AI Summary

This work addresses the limitations of existing approaches to irregular 3D bin packing, which either rely on manually predefined container dimensions or optimize only locally. The paper introduces the first differentiable, end-to-end packing framework that jointly optimizes object poses and all three side lengths of the container. Leveraging axis-aligned bounding box proxies derived from triangle meshes, the method employs a physics-inspired differentiable loss function and an adaptive compression mechanism to dynamically shrink the container—without requiring a physics engine or convex decomposition. Efficient parallel computation is achieved through tensor broadcasting and full-tensor quantization, enabling packing of hundreds of objects in under four minutes on a single consumer-grade GPU. Compared to DBLF and simulated annealing baselines, the approach reduces container volume by 11%–32% and achieves speedups of 3.4× to 54×.

Abstract

Most existing approaches either fix the container in advance or optimize only a single container dimension through an outer search loop, leaving the remaining dimensions as a manual tuning problem. We present a differentiable packing framework that jointly optimizes all 6N object pose parameters and all three container side lengths inside a single gradient-based loop. The formulation combines six physics-inspired, differentiable loss terms computed directly on triangle meshes through axis-aligned bounding-box proxies. An adaptive squeezing mechanism periodically tightens the container whenever the overlap loss falls below a pair-count-scaled threshold, producing a large initial drop in container volume, followed by small refinements. All pairwise computations are written in tensor-broadcasting form, giving a 3.4 to 54 times speedup over a reference loop-based implementation. The pipeline is implemented in Python and PyTorch, with no physics engine, FFT library, or convex decomposition. On multiple object categories, the method produces containers that are 11 to 32 percent smaller than time-matched DBLF and simulated-annealing baselines at N =100, while running in under 4 minutes per instance on a single consumer GPU.

Problem

• 3D packing
• irregular objects
• container optimization
• differentiable optimization
• object pose

Innovation

• differentiable packing
• adaptive container estimation
• tensor broadcasting
• gradient-based optimization
• irregular 3D objects