Sim-to-Real: Learning Agile Locomotion For Quadruped Robots

TL;DR

This paper introduces a system that uses deep reinforcement learning to automatically develop agile quadruped locomotion policies in simulation, which are then successfully transferred to real robots, reducing manual tuning and expertise needed.

Contribution

The authors present a novel simulation-to-real transfer method that improves physics simulation accuracy and policy robustness for quadruped locomotion.

Findings

Robots achieved successful trotting and galloping in real-world tests.

Simulation improvements led to better transfer success.

Robust policies generalize across different physical perturbations.

Abstract

Designing agile locomotion for quadruped robots often requires extensive expertise and tedious manual tuning. In this paper, we present a system to automate this process by leveraging deep reinforcement learning techniques. Our system can learn quadruped locomotion from scratch using simple reward signals. In addition, users can provide an open loop reference to guide the learning process when more control over the learned gait is needed. The control policies are learned in a physics simulator and then deployed on real robots. In robotics, policies trained in simulation often do not transfer to the real world. We narrow this reality gap by improving the physics simulator and learning robust policies. We improve the simulation using system identification, developing an accurate actuator model and simulating latency. We learn robust controllers by randomizing the physical environments, adding perturbations and designing a compact observation space. We evaluate our system on two agile locomotion gaits: trotting and galloping. After learning in simulation, a quadruped robot can successfully perform both gaits in the real world.