Reproduce “Smoothly Differentiable and Efficiently Vectorizable Contact Manifold Generation” 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
Onur Beker, Andreas René Geist, Anselm Paulus, Nico Gürtler, Ji Shi, Sylvain Calinon, Georg Martius
This work introduces a method for generating smoothly differentiable and efficiently vectorizable contact manifolds for rigid-body dynamics simulation in robotics. The framework produces signed distances and normals, achieving orders of magnitude faster vectorized collision detection and gradient computation compared to established simulators like MuJoCo XLA.
Key insights
- An analytical, branchless active set method, when combined with smooth approximations for all non-smooth operations, can significantly outperform iterative interior-point methods in speed and vectorization suitability for small quadratic programs in collision detection.
- Representing surfaces with smooth SDFs derived from primitive combinations allows for direct, differentiable calculation of signed distances and normals from arbitrary points to the surface.
- Smooth activity indicators, integrating collision, penetration, and normal alignment conditions, enable gradual activation and deactivation of contacts, improving the stability of downstream dynamics solvers.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.