Contact Physical Model (CPM) Overview

• CPM is a framework that mathematically formalizes contact interactions and non-penetration for dynamic simulation.
• It employs methods like complementarity constraints, penalty techniques, and smooth surrogates to accurately model friction, adhesion, and deformable contacts.
• CPMs enable contact-implicit trajectory optimization, real-time MPC, and reduced-order modeling, impacting robotics and simulation.

A Contact Physical Model (CPM) formalizes the mechanics of contact between bodies—rigid or deformable—at the level needed for dynamic simulation, optimization, or control. CPMs provide the mathematical and algorithmic framework that governs the enforcement of non-penetration, models the exchange of contact forces (including friction and adhesion), and integrates these constraints into system dynamics or optimization pipelines. They are foundational for contact-implicit trajectory optimization, real-time model predictive control (MPC), deformable body simulation, and model order reduction in contact-rich environments.

1. Mathematical Foundations

CPMs encode the physical interaction between objects using a combination of complementarity conditions, penalty methods, and/or smooth surrogates for contact. The canonical rigid-contact model utilizes the complementarity relation:

• Normal force: $0 \leq \lambda_n \perp \phi(q) \geq 0$, where $\lambda_n$ is the normal contact force and $\phi(q)$ is the gap function.
• Frictional contact (Coulomb): $\|\lambda_t\| \leq \mu \lambda_n$, with additional “stick/slip” selection via complementarity or maximum-dissipation principles, and $\lambda_t$ the tangential force components.

In the context of finite element discretization for contact between deformable bodies, the semi-discrete system reads:

$$\begin{aligned} M\ddot{u}(t) + K u(t) &= f(t) + B^\top \lambda(t) \ g(t) := Bu(t) - c &\geq 0, \;\; \lambda(t) \geq 0, \;\; \lambda(t) \odot g(t) = 0 \end{aligned}$$

where $M$ is the mass matrix, $K$ is stiffness, $B$ extracts normal displacements at contact nodes, $u$ is the displacement vector, $\lambda$ are Lagrange multipliers (contact forces), and $c$ encodes initial clearances (Balajewicz et al., 2015).

Contact-implicit MPC frameworks discretize and linearize these models, casting contact as a Linear Complementarity System (LCS) over a finite time horizon:

$$x_{k+1} = A x_k + B u_k + D \lambda_k + d, \ 0 \leq \lambda_k \perp E x_k + F \lambda_k + H u_k + c \geq 0$$

where $x$ are state variables and $\lambda$ collects normal and frictional contact forces (Aydinoglu et al., 2023, Aydinoglu et al., 2021).

2. Variants and Modeling Paradigms

The literature distinguishes three major CPM categories in computational optimization for manipulation (Onol et al., 2018):

• Complementarity Constraint CPM (CCCM): Directly enforces non-penetration and non-negative contact force via complementarity constraints—exact for rigid contact, but introduces nonconvexity and may hamper convergence in trajectory optimization. The use of a slack variable penalizes small violations for numerical tractability.
• Smooth Contact Model (SCM): Approximates contact forces with parametric smooth surrogates, typically exponential springs, $\gamma(q) = k \exp(-\alpha \phi(q))$, for enhanced differentiability and smooth optimization at the expense of physical realism.
• Variable Smooth Contact Model (VSCM): Introduces adaptivity by optimizing the contact stiffness $k$, interpolating smoothly between “soft” and “hard” contact, providing robust contact discovery and high motion accuracy.

Table: Comparative performance in a pushing task (Onol et al., 2018)

Model	Physical Inaccuracy Φ (N·s)	Final Pos. Error (m)	Solver Time (s)
CCCM	0.1121 – 0.0373	0.1052 – 0.1924	43 – 113
SCM	2.0400 – 0.0381	0.0285 – 0.1908	24 – 41
VSCM	1.2137 – 0.0001	0.0185 – 0.0836	16 – 111

CCCM offers best physical fidelity (lowest virtual forces at small distances), SCM converges fastest but can violate physics by generating “force from a distance,” and VSCM achieves best task accuracy and robustness across varying initial conditions.

3. Computational Methods and Solver Integration

CPMs are embedded in a variety of computational schemes:

• ADMM-based Splitting: In multi-contact MPC, consensus ADMM approaches split the hybrid contact mode selection (via LCP or MIQP projections) from the continuous dynamics, enabling parallelization and real-time performance. Each time step’s contact set is resolved as a small, independent (possibly mixed-integer) projection, decoupling contact “mode scheduling” from the main optimization (Aydinoglu et al., 2023, Aydinoglu et al., 2021).
• Nonlinear Complementarity Problem (NCP) Formulation: For physically accurate frictional contact in deformable simulations (e.g., MPM), the global frictional contact problem is formulated as an NCP over impulses, solved efficiently by ADMM. Contact is enforced via the Signorini-Coulomb laws:

$$\lambda_{n} \geq 0,\; \sigma_{n} \geq 0, \; \lambda_{n}\sigma_{n}=0;\qquad \|\lambda_T\| \leq \mu \lambda_n$$

The variational form minimizes contact energy subject to friction cone constraints (Ménager et al., 2 Feb 2026).

• Projection-based Model Reduction: Reduced-order models for large-scale FE contact problems project the full CPM onto low-dimensional bases for displacements (via POD/SVD) and contact forces (via Non-Negative Matrix Factorization, NNMF), with a greedy sampling and a posteriori error estimation over parameter ranges (Balajewicz et al., 2015).

4. Estimation, Adaptation, and Online Identification

Compliant and differentiable CPMs are central to online estimation and control. A notable approach formulates compliant contact as quadratic penalties on deformation, parameterizing rest positions, locations, and vector stiffness, yielding fully differentiable models suitable for:

• Offline parameter regression via batch optimization,
• Online extended Kalman filtering (EKF) for sensorless force and stiffness estimation,
• Gradient-based online MPC with differentiated contact model and real-time feedback (Haninger et al., 2023).

Experimental validation demonstrates up to 70% force RMSE reduction and ~50% position tracking improvement in contact-rich manipulation tasks by using estimated contact parameters.

5. Application Domains and Performance Benchmarks

CPMs underpin efficient and robust modeling in:

• Trajectory optimization for manipulation and locomotion: Contact-implicit formulations enable direct optimization over contact switches, improving motion quality and accuracy compared to pre-scheduling.
• Multi-contact robot control: Consensus-ADMM MPC architectures achieve real-time planning for high-DOF manipulation with friction and multiple active contacts, with hardware-demonstrated tracking rates up to 240 Hz and feasible schedules for making/breaking contact without prior mode enumeration (Aydinoglu et al., 2023).
• Deformable body and MPM-based simulation: Particle-centric, sub-cell contact localization, combined with NCP+ADMM frictional resolution integrating directly into the implicit MPM step, achieves sub-mm accuracy, robust stick/slip transitions, and stable compaction/plasticity across a range of geometries and materials (Ménager et al., 2 Feb 2026).
• Reduced-order modeling for FE contact problems: Orders-of-magnitude speedups (≥1000×) on both static and dynamic benchmarks, with <1% error, are reported via projection and NNMF basis strategies (Balajewicz et al., 2015).

6. Modeling Assumptions, Regularization, and Limitations

Common CPM simplifications include:

• Rigid body and instantaneous, inelastic impacts: Linearization about the operating point may neglect global geometric nonlinearities (Aydinoglu et al., 2021).
• Contact friction cone approximations: Polyhedral or $\ell_1$-norm approximations of $\ell_2$/Coulomb cones are often used for efficiency.
• Penalty regularization and slack variables: Softening hard constraints for solver stability can introduce virtual forces or small penetrations, with tradeoffs in physical realism and convergence speed (Onol et al., 2018).
• Global uniqueness and solvability: LCP/NCP formulations assume unique solutions per timestep; branching and non-uniqueness are generally unresolved.

7. Future Trends and Comparative Perspectives

Recent CPM advances emphasize:

• Differentiability and parameter identification: Push toward online learning of CPM parameters via end-to-end differentiable pipelines permits closed-loop adaptation to unknown or varying contact properties (Haninger et al., 2023).
• Algorithmic scalability and real-time feasibility: ADMM-based consensus and projection approaches transform previously intractable (MIQP) MPC into practical, parallelizable workloads for high-rate robotic control (Aydinoglu et al., 2023).
• Versatility across simulation and model reduction: Integration into both high-fidelity (MPM, FE) and reduced-order (ROM) environments, with direct handling of frictional, elastic-plastic, and compliant contact, demonstrates the breadth and criticality of CPMs in scientific computing (Ménager et al., 2 Feb 2026, Balajewicz et al., 2015).

CPMs continue to serve as the algorithmic backbone for robust, generalizable, and physically grounded simulation, estimation, and control in contact-rich systems.