Reproduce “YASPS: A Symbolic Framework for Extensible, High-Performance IPC Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xuan Tang, Kemeng Huang, Gilbert Bernstein, Minchen LI, Tzumao Li
YASPS provides a symbolic framework for incremental potential contact simulation. JOIN gathers connected values, while UNION retains multiple vertex parameterizations in one differentiable program. In a point–point barrier test, optimization reduces PSD projection from 24×24 to 6×6. In one 200-frame benchmark, YASPS runs faster overall than GIPC, though differentiation and assembly take longer; collision detection and Newton stepping remain external.
Key insights
- Making connectivity and parameterization explicit in the computation graph avoids enumerating every combination of geometric representation; adding affine-body support required about 130 lines of Python, while a caged-bunny model required about 140 lines without backend changes.
- Symbolic differentiation can remain competitive with hand-written derivatives when the framework exploits shared intermediate expressions, reduced-space projection, and parameterization-aware sparsity rather than treating every local Hessian as dense.
- The same abstractions that improve extensibility introduce practical costs, including large compile times, memory overhead from UNION branches, and dependence on external collision detection and GPU-specific compilation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.