friction characterization

Friction modeling mismatch in robotic simulation constitutes a critical bottleneck underlying the sim-to-real performance gap. This paper proposes a physics-informed neural network (PINN)-based framework for transferable friction estimation, enabling, for the first time, end-to-end differentiable identification of LuGre dynamic friction model parameters. By synergistically integrating LuGre’s physical priors with few-shot, noise-robust data-driven learning, the method accurately reconstructs nonlinear, underactuated system dynamics using only minimal noisy sensor measurements. Unlike heuristic friction models in conventional simulators (e.g., MuJoCo, PyBullet), our approach ensures both physical interpretability and strong generalization—supporting cross-system parameter transfer without retraining. Experimental validation on real hardware demonstrates substantial reduction in sim-to-real discrepancies across diverse robotic platforms, thereby bridging the fidelity gap between simulation and physical deployment.

Existing servo actuator simulations employ coarse friction models (e.g., classical Coulomb-viscous approximations), leading to poor sim-to-real transfer. This work proposes an extended friction model that systematically integrates the Stribeck effect, LuGre hysteresis, and dynamic state evolution—marking the first unified formulation enabling fully automated, trajectory-driven parameter identification from real pendulum-arm motion data. The model features a plug-and-play interface compatible with mainstream physics engines. Evaluated on four servo motors and a 2R manipulator, it reduces mean position and velocity prediction errors by 62% over baseline models and significantly improves cross-domain control policy transfer success. This work establishes a reproducible, deployable friction modeling paradigm for high-fidelity robotic dynamics simulation.

Modern dynamic friction models (DFMs) suffer from poor accessibility and reproducibility due to their structural complexity and lack of intuitive, standardized visual representations. Method: This paper introduces, for the first time, a standardized state-transition block diagram for the generalized Maxwell-slip (GMS) multistate friction model. The diagram uniformly captures state-variable evolution and switching logic, enabling direct implementation in Stateflow or embedded conditional logic, and supporting both open-loop and closed-loop simulation architectures. Contribution/Results: The proposed framework restores the interpretability of classical DFMs, significantly enhancing model comprehension, reproducibility, and engineering deployment efficiency. Simulation results confirm accurate reproduction of non-drift behavior and stick-slip phenomena, and benchmarking against the LuGre model demonstrates high fidelity and practical utility.

This work addresses the limitations of existing rate-dependent friction models, which are often empirical, lack physical interpretability, and fail to satisfy mathematical properties essential for control and estimation. By leveraging fundamental physical principles and inverting the dynamics of bristle elements, the authors propose a physically grounded first-order dynamic friction modeling framework that guarantees stability and passivity. The framework not only recovers lumped-parameter models akin to LuGre but also, for the first time, yields a distributed-parameter hyperbolic partial differential equation (PDE) model directly linked to bristle dynamics, suitable for rolling contact scenarios. Experimental validation demonstrates that the proposed model reproduces key behaviors of the LuGre model while revealing critical differences, thereby exhibiting superior physical consistency and modeling efficacy.

To address performance degradation of data-driven control for nonlinear dynamical systems under high-noise measurements, this paper proposes a robust closed-loop control framework integrating physical priors with machine learning. Our method innovatively embeds, for the first time, a control-aware physics-informed neural network (PINN)—i.e., a PINN explicitly incorporating control inputs—into a model predictive control (MPC) architecture, enabling joint noise-robust dynamics modeling and real-time optimal control. We validate the approach on two high-noise nonlinear benchmarks: the Lorenz-3 chaotic system and a turning lathe. Results show a 42% reduction in modeling error and a 3.1× improvement in closed-loop stability over purely data-driven baselines. The core contribution is the development of the first differentiable, MPC-embeddable, control-aware PINN paradigm, which simultaneously ensures physical consistency, noise robustness, and real-time control performance.

This work addresses the challenge of modeling tendon-driven continuum robots, whose dynamics are highly nonlinear, high-dimensional, and dominated by friction. To overcome these difficulties, a data-driven system identification approach is proposed, integrating the N4SID, ARX, and SINDYc algorithms to construct a low-dimensional dynamical model. The study reveals that the dynamic behavior of multi-segment continuum robots can be accurately captured by a two-degree-of-freedom model, uncovering strong kinematic coupling among joints and substantially reducing modeling complexity. Experimental validation demonstrates that the resulting model achieves high prediction accuracy and has been successfully implemented within a real-time model predictive controller, confirming its effectiveness and practical utility.

This work addresses the scalability challenge in predicting friction coefficients for material pairs, where conventional approaches require exhaustive pairwise experiments whose cost grows quadratically with the number of materials. To overcome this limitation, the authors propose a proxy-interaction-based material embedding framework that learns representations of target materials through interactions with a small set of carefully selected proxy materials. A fusion function then predicts friction coefficients for arbitrary material pairs from their embeddings. The method integrates deterministic and probabilistic mappings, an optimized proxy selection strategy, and a robust mechanism for handling missing data, combining deep learning with physical priors. Evaluated on both simulated and real-world datasets, the approach achieves high prediction accuracy with substantially reduced experimental effort, maintains robustness under partial observability and noise, and enables interpretable embeddings alongside calibrated uncertainty quantification.

This study addresses the unclear role of frictional anisotropy in the locomotion efficiency of biomimetic snake-like robots across diverse terrains. The authors propose a modular, articulated biomimetic snake skin with rapidly interchangeable scales featuring adjustable angles of attack (15°–45°). A custom friction-testing apparatus and a snake robot platform were developed to simultaneously measure multidirectional friction coefficients—including lateral friction—and forward speed on various surfaces such as grass, tree bark, carpet, and smooth substrates. Experimental results reveal systematic variations in frictional properties with scale angle of attack; however, no consistent correlation was found between friction ratio and locomotion speed. This finding suggests that friction ratio alone is insufficient for predicting robotic performance, offering new insights for the design of biomimetic snake skins.

This study addresses the challenges of conventional independently actuated joints in lower-limb exoskeletons—namely, their mechanical complexity, excessive weight, and reliance on torque sensors—by proposing a cable-driven differential architecture tailored for hip–knee flexion–extension movements. The design employs two motors coupled with a linear differential mapping to enable coordinated torque distribution across joints. Combined with a model-based friction compensation strategy, this approach achieves, for the first time in a differential actuation module, high-precision joint torque estimation without the need for physical torque sensors. Experimental validation demonstrates that the proposed method substantially reduces system complexity and mass, offering an effective sensorless torque control solution for lightweight exoskeletons.

This work addresses the challenge of high-precision joint estimation of robotic states and physical parameters under constraints and disturbances by proposing a Bayesian system identification framework. The approach embeds physically consistent inverse dynamics, contact and closed-loop kinematic constraints, and a comprehensive joint friction model as hard equality constraints within the estimation process. Energy-based regressors are incorporated to enhance parameter observability, while prior constraints on inertial and actuation parameters are explicitly supported. A constrained Riccati recursion algorithm preserving banded matrix structure is devised to achieve linear time complexity for efficient computation. Evaluations in simulation and experiments on the Unitree B1 robot demonstrate that, compared to forward-dynamics and decoupled estimation methods, the proposed framework significantly accelerates convergence, reduces errors in inertial and friction parameter estimates, improves contact consistency, and substantially enhances trajectory tracking performance in complex terrains when integrated into model predictive control (MPC).