GainAdaptor: Learning Quadrupedal Locomotion with Dual Actors for Adaptable and Energy-Efficient Walking on Various Terrains

TL;DR

The paper addresses the challenge of energy-efficient, terrain-adaptive quadrupedal locomotion under DRL by introducing GainAdaptor, a dual-actor framework that adaptively tunes joint PD gains while also controlling joint positions. By dividing the action space into PD-gain adjustments and target-position commands, GainAdaptor learns stable, energy-conscious walking across diverse terrains, aided by terrain-aware estimators and a terrain classifier. Empirical results on a Unitree Go1 show substantial energy savings (e.g., up to ~33% compared to open baselines), reduced torque variance, and successful zero-shot tasks like rock climbing, demonstrating strong sim-to-real transfer and real-world robustness. The framework advances practical legged robotics by combining adaptive gains, terrain state estimation, and efficient learning to achieve resilient, energy-aware locomotion in unstructured environments.

Abstract

Deep reinforcement learning (DRL) has emerged as an innovative solution for controlling legged robots in challenging environments using minimalist architectures. Traditional control methods for legged robots, such as inverse dynamics, either directly manage joint torques or use proportional-derivative (PD) controllers to regulate joint positions at a higher level. In case of DRL, direct torque control presents significant challenges, leading to a preference for joint position control. However, this approach necessitates careful adjustment of joint PD gains, which can limit both adaptability and efficiency. In this paper, we propose GainAdaptor, an adaptive gain control framework that autonomously tunes joint PD gains to enhance terrain adaptability and energy efficiency. The framework employs a dual-actor algorithm to dynamically adjust the PD gains based on varying ground conditions. By utilizing a divided action space, GainAdaptor efficiently learns stable and energy-efficient locomotion. We validate the effectiveness of the proposed method through experiments conducted on a Unitree Go1 robot, demonstrating improved locomotion performance across diverse terrains.

Figures (6)

• Figure 1: Adaptive quadrupedal locomotion with optimized joint PD gains. The proposed GainAdaptor framework enables quadrupedal robots to dynamically adjust joint PD gains in response to changing environments, resulting in energy-efficient locomotion. This framework is demonstrated using a Unitree Go1 robot.
• Figure 2: Progression of P and D gains during the training process. The GainAdaptor framework dynamically adjusts the P and D gains based on the terrain conditions.
• Figure 3: Indoor experiments on various terrains.
• Figure 4: Comparison of mean joint torque on level ground.
• Figure 5: Long-range navigation course overview and representative terrains. The left image shows the aerial view of the long-range navigation course marked in red, with segments labeled A through E. The right images show representative examples of the robot traversing different terrains.