# CTS-MoE联合训练感知门控专家实现足式机器人平滑地形过渡

> 原文链接：https://robotics.hellobike.com/papers/2026-11-06-2606-19633-3fc45baa
> 频道：papers | 更新日期：2026-11-06 | 标签：auto, arxiv
> 核心摘要：提出CTS-MoE框架解决多地形运动切换抖动问题，通过联合训练感知门控混合专家与任务特定评论家，在Unitree Go1上实现93.3%连续课程成功率，动作加加速度降低约三分之一，无需地形标签即可仅凭板载深度相机完成户外部署。

# CTS-MoE联合训练感知门控专家实现足式机器人平滑地形过渡

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[Keeping Jerk in Check: Jointly Trained Perception-Gated Experts for Smooth Terrain Transitions](https://arxiv.org/abs/2606.19633)（arXiv 2606.19633）

## 核心内容

针对足式机器人在非连续地形运动中策略切换导致的瞬态抖动与性能下降问题，研究团队提出CTS-MoE（Continuously Transiting Soft Mixture-of-Experts）框架。该方法将多地形运动建模为多任务强化学习，摒弃了传统分层选择器或单体策略蒸馏范式。实验显示，CTS-MoE在连接课程测试中成功率达93.3%，显著优于分层基线（73.3%）和DAgger蒸馏基线（53.3%），并将全程平均动作加加速度（jerk）降低约33%，有效消除了硬切换带来的边界瞬态冲击。

技术架构采用感知门控的稠密混合专家Actor配合任务特定Critic，通过软路由机制实现专家知识的连续组合，同时利用独立Critic隔离异构目标以保留联合优化中的专业化能力。系统引入并发Teacher-Student框架，将特权地形信息下的组合策略迁移至板载深度感知，部署阶段无需地形标签、显式选择器或额外蒸馏。实机验证基于`Unitree Go1`平台，在扩展户外课程中取得10/10全成功率，并成功泛化至四种户外地形类型，验证了纯视觉感知驱动的平滑过渡能力。

## 关键看点

- **软路由组合机制**：采用感知门控稠密MoE Actor替代硬切换，实现专家知识连续插值，消除地形边界处的控制瞬态抖动
- **多任务解耦优化**：配置任务特定Critic隔离异构奖励目标，确保联合训练过程中各专家策略的专业化特性不被同质化
- **无标签视觉部署**：通过并发Teacher-Student架构将特权信息迁移至板载深度图，推理端仅依赖感知输入，无需地形分类标签
- **实机量化验证**：在Unitree Go1上连续课程成功率提升至93.3%，动作Jerk降低约33%，户外四种地形泛化测试10次全胜

## 工程与产业观察

该工作证明了在多地形足式运动中，联合训练软混合专家优于传统的“分治+切换”或“后训练蒸馏”范式。其核心价值在于通过端到端优化解决了感知与控制割裂导致的边界不平滑问题，为具身智能在复杂非结构化环境中的高动态、低冲击运动提供了可复现的算法基准，对提升四足机器人户外作业的稳定性与硬件寿命具有直接工程参考价值。

## 英文摘要 (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: https://cts-moe.github.io/ .

作者：Francisco Affonso, Matheus P. Angarola, Ana Luiza Mineiro, Aditya Potnis, Marcelo Becker, Girish Chowdhary
:::

## 原始参考来源 (Sources)

- [arXiv 2606.19633：Keeping Jerk in Check: Jointly Trained Perception-Gated Experts for Smooth Terrain Transitions](https://arxiv.org/abs/2606.19633)