Extended Friction Models for the Physics Simulation of Servo Actuators

TL;DR

This paper tackles the problem of limited realism in servo-actuator friction modeling, which hampers transferring learned policies from simulation to real robots. It introduces a family of extended friction models that capture Stribeck effects, load dependence, directional behavior, and quadratic friction, and provides a parameter-identification pipeline using pendulum bench trajectories via CMA-ES. The authors integrate these models into servo-actuator and physics-engine simulations, and validate them on four actuators and two 2R arms, showing substantial reductions in MAE compared to the Coulomb-Viscous baseline. The work advances realistic robotic simulation and has practical implications for improving the reliability of RL-based control in real-world robotic systems.

Abstract

Accurate physical simulation is crucial for the development and validation of control algorithms in robotic systems. Recent works in Reinforcement Learning (RL) take notably advantage of extensive simulations to produce efficient robot control. State-of-the-art servo actuator models generally fail at capturing the complex friction dynamics of these systems. This limits the transferability of simulated behaviors to real-world applications. In this work, we present extended friction models that allow to more accurately simulate servo actuator dynamics. We propose a comprehensive analysis of various friction models, present a method for identifying model parameters using recorded trajectories from a pendulum test bench, and demonstrate how these models can be integrated into physics engines. The proposed friction models are validated on four distinct servo actuators and tested on 2R manipulators, showing significant improvements in accuracy over the standard Coulomb-Viscous model. Our results highlight the importance of considering advanced friction effects in the simulation of servo actuators to enhance the realism and reliability of robotic simulations.

Figures (7)

• Figure 1: Comparison on a 2R arm of the Coulomb-Viscous model ($\mathcal{M}_1$), classically used in physics simulators, and a proposed servo actuator model ($\mathcal{M}_4$). The 2R arm (A.) is composed of Dynamixel MX-106 and MX-64 controlled with low gains to mimic the control modes typically used in RL applications. A triangular wave path is tracked by the real system and simulated using two different friction models. The simulated and measured trajectories are presented in (B.), and the error between simulation and measure over time in (C.), highlighting the importance of accounting for various friction effects in the simulation of servo actuators.
• Figure 2: Drive/backdrive diagram for the Coulomb-Viscous model. The static area (gray) is where the system verify $\tau_m + \tau_e + \tau_f = 0$. In the drive (blue) and backdrive (red) areas $|\tau_f| = \tau_f^m$. On the blue and red lines, $\tau_f^m=|\tau_e+\tau_m|$. The dashed line corresponds to configurations where the system is at equilibrium without the need for friction to act ($\tau_m = -\tau_e$).
• Figure 3: On the top, the experiment used to find drive and backdrive torques. Below, the drive/backdrive diagrams obtained for the MX-106 (bottom left) and eRob80:100 (bottom right) servo actuators.
• Figure 4: Drive/backdrive diagrams for the different proposed models, with optimal parameters fitted during the eRob80:100 servo actuator identification. The lines with lower opacity denotes the effect of velocity (1 rad/s per step).
• Figure 5: MAE obtained on the validation logs after identification for each model on the four servo actuators.