State Estimation Of A Wild Robot Toward Validation Of Rigid Body Simulation

There exist few objective measures to evaluate or compare multi-rigid body dynamics simulators. This absence creates uncertainty in simulation capabilities and accuracy. Simulation science has used theory and other simulations (verification) and real-world data (validation) to evaluate simulation correctness. With respect to rigid body dynamics, ballistic rigid body motion has been validated, but simulations involving contact and friction frequently seem to produce results that appear inconsistent with real-world observations; accurate validation has been seldom performed for contacting "rigid" bodies, likely because the observation problem is so challenging (compared to, e.g., fluid dynamics). This paper concentrates on a simple validation scenario for multi-rigid body dynamics with contact and friction, which are essential for simulating robotic locomotion and manipulation. We study collection and estimation of motion data from a mechanically simple but highly dynamic, real-world robot whose motion is primarily driven by contact and friction.