Sizhe Wei, Xulin Chen, Fengze Xie, Garrett Ethan Katz, Zhenyu Gan, Lu Gan
原文摘要
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.
Deep reinforcement learning (DRL) agents are vulnerable to perturbations of their observed state inputs. Prior works often constrain state adversaries with $L_p$-norm uncertainty sets and evaluate policy performance under the resulting perturbations. However, in continuous control tasks like robotics control, different state dimensions have distinct physical meanings and influence policy behavior to different degrees, which $L_p$-norm sets fail to capture or explain. Moreover, it is important to identify when and along which state dimensions a policy is vulnerable, independently of the algorithm used to train it. In this paper, we focus on evaluation-time attacks applied to pretrained policies. We propose \textbf{P}erformance-\textbf{D}egradation \textbf{N}on-\textbf{U}niform \textbf{U}ncertainty \textbf{S}et (\method{}), an interpretable formulation of uncertainty sets for state perturbations. \method{} is state-dependent, dimension-wise non-uniform, and can be synthesized online during policy evaluation. We first formulate an ideal constrained optimization problem for \method{} computation, then introduce a surrogate constraint to make the problem tractable. Experiments across multiple continuous-control environments, policy algorithms, and state attacks show that \method{} reveals variation in policy sensitivity across states and dimensions, while its performance degradation is moderately controlled.