# WM-LOCO世界模型增强视觉步态，Unitree G1零样本迁移成功率92.5%

> 原文链接：https://robotics.hellobike.com/papers/2026-11-04-2609-02542-73d6ce65
> 频道：papers | 更新日期：2026-11-04 | 标签：auto, arxiv
> 核心摘要：针对落脚点受限地形，WM-LOCO利用RSSM世界模型为PPO策略提供预测上下文。仿真平均成功率超94%，并在Unitree G1上实现零样本Sim-to-Real迁移，四类复杂地形综合成功率达92.5%。

# WM-LOCO世界模型增强视觉步态，Unitree G1零样本迁移成功率92.5%

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain](https://arxiv.org/abs/2609.02542)（arXiv 2609.02542）

## 核心内容

针对缝隙、踏石、楼梯及平衡木等落脚点受限地形易导致人形机器人跌倒的难题，研究提出World-Model-Augmented Visual Locomotion（WM-LOCO）框架。该方法通过联合训练的世界模型为视觉运动策略注入预测性上下文，显著提升足端落点规划精度。实验显示，WM-LOCO在仿真中于缝隙、踏石、楼梯和平衡木场景的平均成功率分别达96.0%、96.9%、94.0%和92.1%，全面优于基线方法并改善了步态平滑度与效率。

技术架构采用循环状态空间模型（RSSM）学习动作条件潜在动力学，基于机载深度观测与本体感知预测未来本体状态、自我中心视角观测及任务奖励。生成的循环特征被输入近端策略优化（PPO）Actor-Critic网络，赋予策略对地形的预判能力以指导精确落脚。该系统成功实现零样本Sim-to-Real迁移至Unitree G1人形机器人，在上述四类物理地形测试中取得92.5%的综合成功率，验证了世界模型辅助视觉导航在实体部署中的有效性。

## 关键看点

- **核心架构**：集成RSSM世界模型与PPO算法，利用动作条件潜在动力学预测未来本体状态与视觉观测，为策略提供前瞻上下文
- **仿真性能**：在缝隙、踏石、楼梯及平衡木四类受限地形中，平均成功率介于92.1%至96.9%之间，显著超越现有基线
- **真机验证**：零样本迁移至Unitree G1人形机器人，在四类物理地形实测综合成功率达92.5%，证实Sim-to-Real泛化能力
- **感知输入**：仅依赖机载深度相机与本体感知数据，无需高精地图或外部定位系统即可实现复杂地形下的精准足端规划

## 工程与产业观察

WM-LOCO证明了将世界模型作为“认知插件”嵌入强化学习策略的有效性，解决了纯反应式策略在稀疏支撑面地形下缺乏前瞻性的痛点。其在Unitree G1上的高成功率零样本迁移，表明RSSM学到的潜在动力学具有良好的现实对齐度，为人形机器人在非结构化环境中的鲁棒行走提供了可复现的算法范式与工程参考。

## 英文摘要 (Abstract)

Foothold-constrained terrain limits feasible footholds to sparse, discontinuous, or narrow support regions, as in gaps, stepping stones, stairs, and balance beams. In such settings, a single misplaced step can cause a fall. Reliable traversal therefore requires precise foot placement and anticipation of upcoming terrain. To address this challenge, we present World-Model-Augmented Visual Locomotion (WM-LOCO), which augments a visual locomotion policy with predictive context from a jointly trained world model. Given onboard depth observations and proprioception, a recurrent state-space model (RSSM) learns action-conditioned latent dynamics that predict future proprioceptive states, egocentric observations, and task rewards. The resulting recurrent feature is then fed into the proximal policy optimization (PPO) actor-critic, providing predictive context for anticipatory foot placement on foothold-constrained terrain. WM-LOCO achieves average success rates of 96.0% on gaps, 96.9% on stepping stones, 94.0% on stairs, and 92.1% on balance beams, outperforming the evaluated baselines while improving gait efficiency and smoothness. Furthermore, WM-LOCO achieves zero-shot sim-to-real transfer to a Unitree G1, with 92.5% success across these four terrain classes.

作者：Yuxi Liu, Lijun Han, Ziming Wang, Ao Zhang, Cong Yang, Wei Sui
:::

## 原始参考来源 (Sources)

- [arXiv 2609.02542：World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain](https://arxiv.org/abs/2609.02542)