McARL:Morphology-Control-Aware Reinforcement Learning for Generalizable Quadrupedal Locomotion

TL;DR

McARL introduces a morphology-conditioned reinforcement learning approach enabling a single policy to generalize and transfer across different quadruped robot morphologies, achieving high-speed locomotion without retraining.

Contribution

The paper presents McARL, a novel morphology-aware RL method that improves transferability and generalization of quadruped locomotion policies across diverse robot morphologies.

Findings

Single policy trained on one robot transfers to others with up to 3.5 m/s without retraining.

Achieves 44-150% higher transfer performance compared to PPO variants.

Successfully generalizes to multiple robot morphologies like A1 and Mini Cheetah.

Abstract

We present Morphology-Control-Aware Reinforcement Learning (McARL), a new approach to overcome challenges of hyperparameter tuning and transfer loss, enabling generalizable locomotion across robot morphologies. We use a morphology-conditioned policy by incorporating a randomized morphology vector, sampled from a defined morphology range, into both the actor and critic networks. This allows the policy to learn parameters that generalize to robots with similar characteristics. We demonstrate that a single policy trained on a Unitree Go1 robot using McARL can be transferred to a different morphology (e.g., Unitree Go2 robot) and can achieve zero-shot transfer velocity of up to 3.5 m/s without retraining or fine-tuning. Moreover, it achieves 6.0 m/s on the training Go1 robot and generalizes to other morphologies like A1 and Mini Cheetah. We also analyze the impact of morphology distance on transfer performance and highlight McARL's advantages over prior approaches. McARL achieves 44-150% higher transfer performance on Go2, Mini Cheetah, and A1 compared to PPO variants.