Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning
Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning
This paper presents a morphological symmetry graph representation for locomotion policy learning, instantiated as MS-PPO on Unitree Go2 and MS-SAC on Unitree G1. The method incorporates permutations and sign changes induced by morphological symmetry into the graph actor and critic, achieving lower tracking errors in both simulation and hardware, and the highest mean simulation returns across three joint-failure types on Go2 among evaluated baselines. It also demonstrates zero-shot sim-to-real deployment on the 29-DoF Unitree G1 whole-body locomotion task.
The authors introduce a morphological symmetry graph representation (MS-PPO/MS-SAC) that explicitly encodes physical symmetries in robot observations and actions, achieving lower tracking errors and higher simulation returns on Unitree Go2 and G1 platforms compared to baselines, with zero-shot sim-to-real deployment demonstrated.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2512.00727v3 Announce Type: replace 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 morphologic
Open source S2
al 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 repr
Open source S3
esentation 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
Open source S4
the PPO baseline. These results support the effectiveness of jointly encoding topology and morphological symmetry across robot embodiments and learning frameworks.
Open source S5
Beyond Topology: A Morphological Symmetry Graph Representation for Locomotion Policy Learning
Open source S1
Source：arXiv Robotics — research abstracts · arxiv.org