True Rigidity: Interpenetration-free Multi-Body Simulation with Polytopic Contact
This addresses accuracy and reliability issues in robot simulation for locomotion and manipulation, though it is incremental as it builds on existing time-stepping methods.
The paper tackles the problem of determining convergence and preventing interpenetration in multi-rigid body simulations with rigid contact, by computing event times for contact/impact events and using a variable integration approach that ensures no interpenetration for convex polytopic geometries.
An effective paradigm for simulating the dynamics of robots that locomote and manipulate is multi-rigid body simulation with rigid contact. This paradigm provides reasonable tradeoffs between accuracy, running time, and simplicity of parameter selection and identification. The Stewart-Trinkle/Anitescu-Potra time stepping approach is the basis of many existing implementations. It successfully treats inconsistent (Painleve-type) contact configurations, efficiently handles many contact events occurring in short time intervals, and provably converges to the solution of the continuous time differential algebraic equations (DAEs) as the integration step size tends to zero. However, there is currently no means to determine when the solution has largely converged, i.e., when smaller integration steps would result in only small increases in accuracy. The present work describes an approach that computes the event times (when the set of active equations in a DAE changes) of all contact/impact events for a multi-body simulation, toward using integration techniques with error control to compute a solution with desired accuracy. We also describe a first-order, variable integration approach that ensures that rigid bodies with convex polytopic geometries never interpenetrate. This approach permits taking large steps when possible and takes small steps when contact is complex.