# CADENCE：强化学习非周期步态规划器四足崎岖地形通过率85.1%

> 原文链接：https://robotics.hellobike.com/papers/2026-11-18-2611-23102-3d343b79
> 频道：papers | 更新日期：2026-11-18 | 标签：auto, arxiv
> 核心摘要：提出CADENCE非周期步态规划器，以40ms频率在65个离散动作中选择落足点与摆动时间。RL策略结合凸MPC训练，无预设步态表，在26%-49%空隙率地形仿真中成功率达85.1%，显著优于OCS2-Trot等传统方法。

# CADENCE：强化学习非周期步态规划器四足崎岖地形通过率85.1%

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[CADENCE: Learning Acyclic Gaits for Model-Based Quadruped Locomotion](https://arxiv.org/abs/2611.23102)（arXiv 2611.23102）

## 核心内容

针对传统模型预测控制（MPC）依赖固定周期步态导致复杂地形适应性差的问题，研究团队提出CADENCE学习型步态规划器。该系统每40ms从最多65个离散选项中决策单腿落足或保持静止，并预测摆动时长。在包含26%至49%空隙率的崎岖地形及0.1-0.2m/s速度仿真测试中，CADENCE五次种子平均成功率达85.1%，远超OCS2-Trot的67.8%和OCS2-Single的65.5%；在49%极端空隙率下仍以72.9%成功率领先基准方案近30个百分点。

技术上，CADENCE通过强化学习在凸MPC之上训练，奖励函数仅含步进成本而无特定步态项，仅约束同时摆动腿数不超过两条，促使策略自发涌现出对角配对的非周期步态。系统引入逆步态模型（IFM）将落足点转化为控制指令，仅需重拟合IFM即可在少于2万步数据下冻结部署于OCS2非线性MPC。消融实验显示，去除动态规划的静态行走变体成功率降至39.6%，验证了学习型非周期规划对高难度地形的关键增益价值。

## 关键看点

- **离散决策机制**：40ms周期内从≤65个离散选项中选择单腿安全落足点及摆动时间，摒弃固定步态时序表
- **非周期步态涌现**：RL奖励仅设步进成本且限制最多两腿同摆，策略自主习得无固定周期的对角腿配对模式
- **仿真性能验证**：26%-49%空隙率地形平均通过率85.1%，49%极端工况达72.9%，较OCS2-Trot提升约30%
- **低样本迁移部署**：利用逆步态模型IFM转换指令，仅需&lt;2万步日志重拟合即可冻结接入OCS2非线性MPC

## 工程与产业观察

该工作证明了在非结构化地形中，基于离散动作空间的RL步态规划可替代手工设计的周期步态表，且能与经典凸MPC无缝耦合。其「策略冻结+IFM重拟合」的低样本部署范式，为四足机器人在缺乏精确动力学模型时的快速现场适配提供了工程参考，降低了具身控制算法从仿真到实机的迁移门槛。

## 英文摘要 (Abstract)

Model-based quadruped controllers usually step on a cyclic gait, which lifts the legs in a fixed order on a fixed period, wherever the safe footholds are. We present CADENCE, a learned step planner with a small action space. Every 40 ms, it picks one of at most 65 discrete choices: no step, or one leg's step to a safe foothold. It also predicts a swing time. We train CADENCE by reinforcement learning on top of a convex model predictive controller (MPC). Nothing prescribes a gait: there is no gait schedule, and the reward has no gait term, only a cost per step. The one structural constraint is that at most two legs may swing at once. The policy we analyzed developed an acyclic gait, with no fixed period or stepping order, yet it still paired diagonal legs. An inverse footstep model (IFM), fitted on a controller's logged steps, turns each foothold into a command. With only the IFM refitted, on fewer than 20,000 steps, CADENCE ran frozen on OCS2's nonlinear MPC. In simulated trials on a range of terrains (26% to 49% voids) and speeds (0.1 m/s and 0.2 m/s), it succeeded in 85.1% (mean of five seeds), against 67.8% for OCS2-Trot, 65.5% for OCS2-Single, and 39.6% in a static walk ablation of CADENCE. At 49% voids, it succeeded in 72.9%, against 43.0% for OCS2-Trot.

作者：Owen Sullivan, Sourav Selvaraj, Mahdi Agheli
:::

## 原始参考来源 (Sources)

- [arXiv 2611.23102：CADENCE: Learning Acyclic Gaits for Model-Based Quadruped Locomotion](https://arxiv.org/abs/2611.23102)