A Tutorial on Complementarity-Based Contact Models for Robotics
Organizations: University of Tübingen · University of Pennsylvania · Technical University Berlin · Inria & DI-ENS
Abstract
Simulating rigid-body dynamics with contact is fundamental to robotics, from motion synthesis and control to mechanism design. Many widely used simulators rely on cone complementarity formulations, whose physical assumptions and numerical solution require substantial background. While excellent surveys, monographs, and course notes exist, they mostly address computational mechanics or computer graphics or document a single simulator. To the best of our knowledge, a self-contained pedagogical treatment for robotics, bringing these formulations and their solvers together with complete derivations and a unified notation, does not yet exist. This tutorial addresses this gap by connecting rigid-body dynamics, conic optimization, and frictional contact. In particular, we explain how complementarity conditions encode non-penetration and frictional dissipation, determining separation, sticking, and sliding without prescribed contact modes. We distinguish equivalent transcriptions from convex relaxations, showing how the latter can introduce artifacts such as lift-off and creep. We examine the existence and uniqueness of motion and contact forces, including the solver's role in selecting among multiple solutions. We then compare first-order and Newton-based methods, explain regularization and proximal strategies, and discuss trade-offs between computational cost, robustness, and physical fidelity. Connections to widely used simulators, optimal control, and differentiable simulation help readers interpret simulation results, choose suitable formulations and solvers, and develop their own contact dynamics routines.
Figures & tables
| Transcriptions | Convex | Solution exists | Unique motion | Unique forces | Physical artifact |
|---|---|---|---|---|---|
| Signorini–Coulomb, N-CCP, root-finding, C-QCQP, C-QP (exact) | no | under a strict feasibility assumption ( Acary et al., 2011 ) | not in general | not in general | none |
| L-CCP, QCQP | yes | if admissible velocities exist, e.g., for | yes | if is nonsingular | sliding contacts lift off |
| Regularized L-CCP and QCQP, NLS | strongly | always | yes | yes | lift-off, creep, compliant normal response |
| LCP ( Appendix A ) | no | always ( Anitescu and Potra, 1997 ) | not in general | not in general | anisotropic friction |
| Algorithm | Transcription | Linear solve per iteration | Convergence | Singular ; large mass ratios | GPU suitability | Warm start | Early termination | Smooth solutions |
| PGS (Alg. 1 ) | L-CCP | none | global; at best linear | robust; slow | Jacobi or coloring variants | yes | feasible forces | no |
| Newton, natural map (Alg. 2 ) | L-CCP | local, superlinear | may be singular; robust | batched solves | yes | may leave | no | |
| PGD (Alg. 3 ) | QCQP | none | global; linear, grows with the condition number | robust; slow | fully parallel | yes | feasible forces | no |
| ADMM (Alg. 4 ) | QCQP | one factorization, reused | global; speed depends on | robust; robust | batched solves | yes | feasible forces | no |
| Interior point (Alg. 5 ) | QCQP | , refactorized | global; few iterations | robust; robust | batched solves | hard | strictly feasible | yes, for |
| Newton (Alg. 6 ) | NLS | , sparse | global; few iterations | robust; robust | batched solves | yes | feasible forces | no |
| Engine | Transcription | Solver | Hardware |
|---|---|---|---|
| MuJoCo ( Todorov, 2014 ; Google DeepMind, 2022 ) ; MJX, MuJoCo Warp ( Google Deepmind, 2025 ; Google DeepMind and NVIDIA, 2025 ) | Regularized QCQP ( 59 ) (dual), NLS ( 60 ) (primal). | Newton (default) or CG on the primal; PGS on the dual. | CPU (MuJoCo); GPU (MJX, MuJoCo Warp) |
| Drake, SAP ( Castro et al., 2023 ; Tedrake and the Drake Team, 2019 ) | NLS ( 60 ). | Newton with exact line search. | CPU |
| Bullet ( Coumans, 2015 ) | C-QP ( 55 ), LCP ( 85 ). | PGS, Pivoting. | CPU |
| ODE ( Smith, 2006 ) ; DART ( Lee et al., 2018 ) | Boxed LCP ( Andrews et al., 2022 ) . | PGS, Pivoting. | CPU |
| PhysX, Isaac ( Makoviychuk et al., 2021 ) | LCP ( 85 ). | PGS, temporal Gauss–Seidel (TGS). | CPU, GPU |
| Chrono ( Tasora et al., 2016 ) | L-CCP/QCQP ( 58 ). | PGS, PGD , ADMM. | CPU |
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Robotics Simulation
Robotics simulation builds virtual robots and environments to model, train, and evaluate autonomous systems before real-world deployment.
See all 351 papers