# 形态对称图表示MS-PPO/SAC实现Go2与G1低参数量运动策略学习

> 原文链接：https://robotics.hellobike.com/papers/2026-11-10-2512-00727-2d9e75ce
> 频道：papers | 更新日期：2026-11-10 | 标签：auto, arxiv
> 核心摘要：提出形态对称图表示法解决RL策略与机器人物理对齐弱的问题，在Unitree Go2和G1上验证。MS-PPO参数量仅为基线34.6%，在指令跟踪与关节故障任务中实现最低误差及最高仿真回报，并支持G1零样本Sim-to-Real部署。

# 形态对称图表示MS-PPO/SAC实现Go2与G1低参数量运动策略学习

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning](https://arxiv.org/abs/2512.00727)（arXiv 2512.00727）

## 核心内容

针对强化学习运动策略与机器人物理结构对齐不足的问题，研究提出形态对称图表示法，显式编码拓扑连接及对称性诱导的观测/动作置换与符号变化。该方法实例化为Go2平台的MS-PPO与G1平台的MS-SAC（FastSAC），构建了具备等变动作均值与不变Critic的图Actor网络，旨在提升策略对机体对称性的利用效率。

实验覆盖Unitree Go2指令跟踪、非对称关节故障及Unitree G1 29自由度全身运动控制与零样本Sim-to-Real部署。结果显示MS-PPO在Go2上仅使用PPO基线34.6%的参数，且在仿真与真机中均取得最低跟踪误差；在三类关节故障测试中获得最高平均仿真回报，证实了联合编码拓扑与形态对称性在跨本体运动学习中的有效性。

## 关键看点

- **参数效率**：MS-PPO在Unitree Go2指令跟踪任务中，仅消耗PPO基线34.6%的参数量即实现更优性能
- **鲁棒性验证**：在Go2三类非对称关节故障任务中取得最高平均仿真回报，显著优于传统通用网络基线
- **Sim-to-Real**：在Unitree G1 29-DoF全身运动控制中成功实现零样本真机部署，验证跨平台泛化能力
- **架构创新**：构建等变动作均值与不变Critic的图Actor，显式处理对称性诱导的观测与动作置换及符号变化

## 工程与产业观察

该工作证明了将机器人形态对称性作为归纳偏置嵌入RL网络，可在大幅降低模型参数量的同时提升运动策略的鲁棒性与真机迁移成功率。对于四足与人形机器人开发者，这为在算力受限端侧部署高性能运动控制器提供了基于几何先验的高效算法范式，减少了对海量数据的依赖。

## 英文摘要 (Abstract)

Reinforcement learning has enabled impressive locomotion skills on articulated robots, but common policy representations remain only weakly aligned with robot physics. Generic networks ignore kinematic structure, while graph-based policies encode connectivity without specifying how physical quantities transform across symmetric body parts. We introduce a morphological symmetry graph representation for locomotion policy learning, instantiated as MS-PPO with PPO on Go2 and MS-SAC with FastSAC on G1. Starting from the robot's topological graph, our representation explicitly incorporates the permutations and sign changes induced by morphological symmetry on observations and actions, yielding a graph actor with an equivariant action mean and an invariant critic. We evaluate the representation on Unitree Go2 command tracking and asymmetric joint-failure tasks, and on Unitree G1 29-DoF whole-body locomotion, including zero-shot sim-to-real deployment. Our approach achieves the lowest tracking errors in both simulation and hardware, and the highest mean simulation returns across all three joint-failure types on Go2 among the evaluated baselines. On Go2, MS-PPO uses only 34.6% of the parameters of the PPO baseline. These results support the effectiveness of jointly encoding topology and morphological symmetry across robot embodiments and learning frameworks.

作者：Sizhe Wei, Xulin Chen, Fengze Xie, Garrett Ethan Katz, Zhenyu Gan, Lu Gan
:::

## 原始参考来源 (Sources)

- [arXiv 2512.00727：Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning](https://arxiv.org/abs/2512.00727)