Group relative return policy improves robot control without critics
GR2PO: Group Relative Return Policy Optimization for Continuous Robot Control
Robotics
Summary
Controlling robots smoothly usually needs training methods that estimate long-term outcomes, which can be slow and complex. The authors propose a new way called GR2PO that avoids those complex estimations by comparing groups of results to guide learning. This method helps robots learn better movements faster without extra computational steps. It also works well when tested both in simulations and on real devices like small computers attached to robots.
What this means in practice
- •For robotics engineers: Train robot controllers more efficiently without needing complex value estimation models in reinforcement learning.
- •For embedded systems developers: Deploy reinforcement learning-based robot control policies on low-power edge devices like NVIDIA Jetson TX2 for real-time operation.
Authors
Pengqin Wang, Qiming Zhang, Shaojie Shen, Jun Ma
Abstract
Actor-critic architecture has been widely used in continuous robot control. However, they rely on learning a value network, introducing additional computational overhead during training. Moreover, policy learning may also be affected by the approximation error of value estimation. Critic-free group relative policy optimization methods provide a simpler training approach by removing the need for a critic. However, they fail to learn long-term action outcomes when directly applying immediate rewards to policy optimization in dense-reward environments. To address these problems, we propose Group Relative Return Policy Optimization (GR2PO), a critic-free reinforcement learning framework for continuous robot control. GR2PO estimates the discounted returns from the parallelly collected trajectories, performs group normalization at each rollout time index, and uses relative advantages and clipped targets to update the policy. To evaluate the effectiveness of the proposed framework, we instantiate it on robot control simulation environments and deploy the model to a real-world edge device. The results show that GR2PO significantly outperforms critic-free baselines that use immediate rewards and performs competitively against state-of-the-art actor-critic methods. Furthermore, GR2PO demonstrates competitive training efficiency. Inference tests on NVIDIA Jetson TX2 demonstrate the feasibility of deploying the learned policies on edge platforms. Further ablation experiments analyze the effects of parallel group size, return estimation methods, and target clipping ratio on learning performance. To support follow-up research, we will make the complete code publicly available after the paper is accepted, including the framework implementation, experimental configuration, and training and evaluation scripts.