Reproduce “Scalable Differentiable Physics for Learning and Control” 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

Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin

The authors develop a differentiable simulator for mesh-based collision scenes. It groups contacts into local impact zones and accelerates gradients through collision optimization. In tested benchmarks, runtime and memory remain lower than ChainQueen as object count or relative scale increases; ChainQueen runs out of memory at 200 objects. For 300 connected cubes, acceleration reduces backpropagation time from 21.88s to 1.30s.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.