Physics-Informed Learning for the Friction Modeling of High-Ratio Harmonic Drives
Ines Sorrentino, Giulio Romualdi, Fabio Bergonti, Giuseppe ĽErario, Silvio Traversaro, Daniele Pucci
TL;DR
The paper tackles friction identification in humanoid joints driven by high-ratio harmonic drives, where precise torque control is hindered by significant friction and lack of joint torque sensors. It presents a Physics-Informed Neural Network (PINN) that uses the history of joint-position error and velocities to infer friction from robot data without dedicated test rigs, guided by the robot's dynamics and motor equations, e.g. $M \dot{\\nu} + h(q,\\nu) = B \\tau + \\sum_k J_k(q)^T f_{ext,k}$ and $k_t i_m = J_m \\ddot{\\theta} + \frac{1}{r} \\tau_F + \frac{1}{r} \\tau$, with $\\tau \\approx r k_t i_m - \\tau_F$ in the friction-dominated regime. The PINN approach is benchmarked against classical static friction models (Coulomb-viscous and Stribeck-Coulomb-viscous) and integrated into a two-layer torque control scheme to achieve real-time friction compensation, with validation on two ergoCub joints showing reduced energy losses while maintaining tracking performance. The method is scalable to large joint counts, enabling robust, energy-efficient control for humanoid robots without extra sensing hardware. Overall, the work demonstrates that physics-informed learning can deliver accurate, computationally efficient friction estimates across many joints, improving control fidelity and efficiency in legged robotics.
Abstract
This paper presents a scalable method for friction identification in robots equipped with electric motors and high-ratio harmonic drives, utilizing Physics-Informed Neural Networks (PINN). This approach eliminates the need for dedicated setups and joint torque sensors by leveraging the roboťs intrinsic model and state data. We present a comprehensive pipeline that includes data acquisition, preprocessing, ground truth generation, and model identification. The effectiveness of the PINN-based friction identification is validated through extensive testing on two different joints of the humanoid robot ergoCub, comparing its performance against traditional static friction models like the Coulomb-viscous and Stribeck-Coulomb-viscous models. Integrating the identified PINN-based friction models into a two-layer torque control architecture enhances real-time friction compensation. The results demonstrate significant improvements in control performance and reductions in energy losses, highlighting the scalability and robustness of the proposed method, also for application across a large number of joints as in the case of humanoid robots.