Assemble Them All

Assembly planning is the core of automating product assembly, maintenance, and recycling for modern industrial manufacturing. Despite its importance long history research, mechanical assemblies when given final assembled state remains a challenging problem. This due to complexity dealing with arbitrary 3D shapes highly constrained motion required real-world assemblies. In this work, we propose novel method efficiently plan physically plausible assembly sequences Our leverages assembly-by-disassembly principle physics-based simulation explore reduced search space. To evaluate generality our method, define large-scale dataset consisting thousands valid variety motions required. experiments on new benchmark demonstrate achieve state-of-the-art success rate highest computational efficiency compared other baseline algorithms. also generalizes rotational (e.g., screws puzzles) solves 80-part within several minutes.