arXiv：cs.AI· Ziyang Cheng, Haoyu Wei, Hang Yin, Xiuwei Xu, Bingyao Yu, Jie Zhou, Jiwen Lu·· 10 小时前
CMP：通过能力流形投影实现鲁棒的全身移动操作跟踪
CMP: Robust Whole-Body Tracking for Loco-Manipulation via Competence Manifold Projection
AI 导读
针对腿式移动机械臂全身控制策略在传感器噪声或不可行指令下易受OOD输入影响的问题，研究提出 Competence Manifold Projection（CMP），通过 Frame-Wise Safety Scheme、Lower-Bounded Safety Estimator 与 Isomorphic Latent Space 实现 O(1) 防御。
正文
Abstract:While decoupled control schemes for legged mobile manipulators have shown robustness, learning holistic whole-body control policies for tracking global end-effector poses remains fragile against Out-of-Distribution (OOD) inputs induced by sensor noise or infeasible user commands. To improve robustness against these perturbations without sacrificing task performance and continuity, we propose Competence Manifold Projection (CMP). Specifically, we utilize a Frame-Wise Safety Scheme that transforms the infinite-horizon safety constraint into a computationally efficient single-step manifold inclusion. To instantiate this competence manifold, we employ a Lower-Bounded Safety Estimator that distinguishes unmastered intentions from the training distribution. We then introduce an Isomorphic Latent Space (ILS) that aligns manifold geometry with safety probability, enabling efficient O(1) seamless defense against arbitrary OOD intents. Experiments demonstrate that CMP achieves up to a 10-fold survival rate improvement in typical OOD scenarios where baselines suffer catastrophic failure, incurring under 10% tracking degradation. Notably, the system exhibits emergent "best-effort" generalization behaviors to progressively accomplish OOD goals by adhering to the competence boundaries. Project website: this https URL
| Comments: | Accepted to CoRL 2026 (Spotlight). Project webpage: this https URL |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2604.07457 [cs.RO] |
| (or arXiv:2604.07457v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2604.07457 arXiv-issued DOI via DataCite |
Submission history
From: Ziyang Cheng [view email]
[v1]
Wed, 8 Apr 2026 18:00:39 UTC (16,698 KB)
[v2]
Mon, 12 Oct 2026 16:50:13 UTC (18,574 KB)
来源：arXiv：cs.AI · arxiv.org