Benefit of Fixed-Part Priors in Assembly Methods

Determine whether fixed-part 3D assembly methods benefit from exploiting the fixed, known part set through super-part relation reasoning or order encoding, in contrast to the more challenging library-based retrieval and assembly setting where that prior is unavailable.

Background

The paper compares PART with methods that assemble a fixed, predefined set of parts. The authors observe that these prior methods may exploit information unavailable in the library-based retrieval setting, including knowledge of the complete part set, super-part relationships, and part ordering. They present this explanation as a conjecture for why those methods perform well despite operating under a less general assembly condition.

The unresolved issue is whether the performance of fixed-part assembly methods is indeed attributable to these specific priors, rather than to other architectural or training differences. Establishing this would clarify the sources of performance advantages in fixed-part versus retrieval-based 3D assembly.