Validating Robotics Simulators on Real-World Impacts

A realistic simulation environment is an essential tool in every roboticist’s toolkit, with uses ranging from planning and control to training policies reinforcement learning. Despite the centrality of modern robotics, little work has been done comparing robotics simulators against real-world data, especially for scenarios involving dynamic motions high speed impact events. Handling contact computational bottleneck most simulations, thus modeling algorithmic choices surrounding impacts friction form largest distinctions between popular tools. Here, we evaluate ability several reproduce trajectories impacts. Using experimental identify system-specific parameters Drake, MuJoCo, Bullet, analyzing effects around these parameters. For simple example a cube tossed onto table, capture inelastic surprisingly well, though generally fail elasticity. higher-dimensional case Cassie biped landing jump, bulk motion well but accuracy limited by model differences real robot simulators.