arXiv cs.AI· Francisco Affonso, Matheus P. Angarola, Ana Luiza Mineiro, Aditya Potnis, Marcelo Becker, Girish Chowdhary·· 11 小时前AI 评分31
CTS-MoE：联合训练感知门控专家实现平滑地形过渡
Keeping Jerk in Check: Jointly Trained Perception-Gated Experts for Smooth Terrain Transitions
AI 导读
CTS-MoE 将多地形运动建模为多任务强化学习，联合训练感知门控的稠密 MoE actor 与任务专属 critic，通过软路由实现连续变化的行为组合。相比分层选择和 DAgger 蒸馏基线，该策略将连通赛道成功率从 73.3% 和 53.3% 提升至 93.3%，全程平均动作 jerk 降低约三分之一，并消除了硬切换在地形边界处的瞬态。
正文 · 原文
Abstract:Perceptive legged locomotion across discontinuous terrain requires shared locomotion knowledge, task-specific maneuvers, and smooth transitions between them. Existing approaches train specialized policies independently and either switch among them with a hierarchical selector or distill them into a monolithic policy. Both fix the constituent behaviors before they are composed, and both are assessed through aggregate course outcomes that leave behavior at terrain boundaries uncharacterized. We instead propose CTS-MoE, which formulates multi-terrain locomotion as multi-task reinforcement learning and jointly trains a perception-gated dense mixture-of-experts actor with task-specific critics: soft routing composes expert knowledge into continuously varying behavior, while separated critics isolate heterogeneous objectives so that specialization survives joint optimization. A concurrent teacher-student formulation transfers this composition from privileged terrain information to onboard depth, so that at deployment routing is inferred from perception alone, with no terrain label, selector, or distillation stage. Against hierarchical and DAgger-distilled baselines, the policy raises connected-course success to 93.3% from 73.3% and 53.3%, lowers mean action jerk by about a third across the course, and avoids the switching transient that hard selection introduces at boundaries. On a Unitree Go1, it completes an extended outdoor course in 10/10 trials and transfers to four outdoor terrain types. Project Website: this https URL .
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2606.19633 [cs.RO] |
| (or arXiv:2606.19633v2 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2606.19633 arXiv-issued DOI via DataCite |
Submission history
From: Francisco Affonso [view email]
[v1]
Wed, 17 Jun 2026 22:25:49 UTC (23,276 KB)
[v2]
Thu, 5 Nov 2026 07:18:21 UTC (22,210 KB)
来源：arXiv cs.AI · arxiv.org