Uploaded March 2023 | Updated September 2026, 3 weeks ago
Abstract:
Robotic locomotion is often approached with the goal of maximizing robustness and reactivity by increasing motion control frequency. We challenge this intuitive notion by demonstrating robust and dynamic locomotion with a learned motion controller executing at as low as 8 Hz on a real ANYmal C quadruped. The robot is able to robustly and repeatably achieve a high heading velocity of 1.5 m/s, traverse uneven terrain, and resist unexpected external perturbations. We further present a comparative analysis of deep reinforcement learning (RL) based motion control policies trained and executed at frequencies ranging from 5 Hz to 200 Hz. We show that low-frequency policies are less sensitive to actuation latencies and variations in system dynamics. This is to the extent that a successful sim-to-real transfer can be performed even without any dynamics randomization or actuation modeling. We support this claim through a set of rigorous empirical evaluations. Moreover, to assist reproducibility, we provide the training and deployment code along with an extended analysis on the project website.
Authors: Siddhant Gangapurwala, Luigi Campanaro and Ioannis Havoutis
Accepted to IEEE International Conference on Robotics and Automation (ICRA) 2023
Project Website: ori-drs.github.io/lfmc
Code:
Training (RaiSim - C++): github.com/ori-drs/lfmc_gym
Deployment (RaiSim - C++): github.com/ori-drs/lfmc_cval
Deployment (PyBullet/RaiSim - Python): github.com/ori-drs/lfmc_pyval
Pre-print: arxiv.org/abs/2209.14887
Dog Clip: pexels.com/video/video-of-dog-playing-with-a-ball-4651948
Abstract:
Robotic locomotion is often approached with the goal of maximizing robustness and reactivity by increasing motion control frequency. We challenge this intuitive notion by demonstrating robust and dynamic locomotion with a learned motion controller executing at as low as 8 Hz on a real ANYmal C quadruped. The robot is able to robustly and repeatably achieve a high heading velocity of 1.5 m/s, traverse uneven terrain, and resist unexpected external perturbations. We further present a comparative analysis of deep reinforcement learning (RL) based motion control policies trained and executed at frequencies ranging from 5 Hz to 200 Hz. We show that low-frequency policies are less sensitive to actuation latencies and variations in system dynamics. This is to the extent that a successful sim-to-real transfer can be performed even without any dynamics randomization or actuation modeling. We support this claim through a set of rigorous empirical evaluations. Moreover, to assist reproducibility, we provide the training and deployment code along with an extended analysis on the project website.
Authors: Siddhant Gangapurwala, Luigi Campanaro and Ioannis Havoutis
Accepted to IEEE International Conference on Robotics and Automation (ICRA) 2023
Project Website: ori-drs.github.io/lfmc
Code:
Training (RaiSim - C++): github.com/ori-drs/lfmc_gym
Deployment (RaiSim - C++): github.com/ori-drs/lfmc_cval
Deployment (PyBullet/RaiSim - Python): github.com/ori-drs/lfmc_pyval
Pre-print: arxiv.org/abs/2209.14887
Dog Clip: pexels.com/video/video-of-dog-playing-with-a-ball-4651948
