(The Eurographics Association, 2024) Nassif, Loïc; Zoubir, O.; Andrews, Sheldon; Kry, Paul G.; Zordan, Victor
We present a novel data-driven approach for simulating friction between rigid bodies that captures the rich diversity of frictional behaviors that arises due to the complex interactions of micro-asperities of different surfaces. Rather than performing detailed simulations with expensive collision detection, we parameterize our friction model based on aggregate features of pairs of surfaces, such as the distribution of normals from each surfaces, which may be easily computed from a texture-based embedding. Our data-driven model is constructed by conducting real-world planar pushing experiments that capture the friction behavior of many different material pairs, and we then fit this data using a Gaussian process (GP). The trained GP model is then evaluated in a real-time simulation and used to update the limit surface used by the contact solver.