Learning Transferable Friction Models and LuGre Identification via Physics Informed Neural Networks

TL;DR

The paper addresses the sim-to-real gap in robotics caused by oversimplified friction models by introducing a physics-informed neural network framework that hybridizes LuGre friction with learnable components. It demonstrates both black-box friction estimators and LuGre parameter estimators that learn from minimal, noisy data while enforcing physical consistency through the equations of motion. The approach yields online estimation capabilities and transferability across systems sharing the same environment, demonstrated on the nonlinear Pendulum-on-a-Box and a Spring-Damper-on-a-Box testbed, with faster parameter identification than full LuGre fitting and accurate reproduction of friction dynamics. This work offers a scalable, interpretable path toward bridging the sim-to-real gap in robotics and control, enabling more reliable underactuated tasks and friction-aware planning.

Abstract

Accurately modeling friction in robotics remains a core challenge, as robotics simulators like Mujoco and PyBullet use simplified friction models or heuristics to balance computational efficiency with accuracy, where these simplifications and approximations can lead to substantial differences between simulated and physical performance. In this paper, we present a physics-informed friction estimation framework that enables the integration of well-established friction models with learnable components-requiring only minimal, generic measurement data. Our approach enforces physical consistency yet retains the flexibility to adapt to real-world complexities. We demonstrate, on an underactuated and nonlinear system, that the learned friction models, trained solely on small and noisy datasets, accurately simulate dynamic friction properties and reduce the sim-to-real gap. Crucially, we show that our approach enables the learned models to be transferable to systems they are not trained on. This ability to generalize across multiple systems streamlines friction modeling for complex, underactuated tasks, offering a scalable and interpretable path toward bridging the sim-to-real gap in robotics and control.

Figures (8)

• Figure 1: Pendulum-on-a-Box system
• Figure 2: I/O of (a) Blackbox Models, (b) Parameter Estimation Models
• Figure 3: BB$_1$ and PE$_1$ friction estimations for: (a) Traj. 1, (b) Traj. 2
• Figure 4: Online estimation performances of BB$_{1,2}$ and PE$_{1,2}$ for Traj. 2.
• Figure 5: Friction force characteristics for (a) PE$_1$, (b) BB$_1$, (c) LuGre model