Single-Level Differentiable Contact Simulation

We present a differentiable formulation of rigid-body contact dynamics for objects and robots represented as compositions convex primitives. Classical physics engines rely on non-differentiable collision detection modules. More recent optimization-based approaches simulating between primitives bilevel that separates simulation. These latest are unreliable in realistic simulation scenarios because isolating the problem introduces location non-uniqueness. Our approach combines into unified single-level optimization problem. This disambiguates physics-informed manner. features improved robustness reduction computational complexity when compared to similar differentiable baseline. illustrate differentiability robotic manipulation task requiring optimization-through-contact. provide numerically efficient implementation our called https://github.com/simon-lc/Silico.jl Silico.jl.