# 迈向基于 MoE 的鲁棒四足运动的可靠 Sim-to-Real 可预测性

> 原文链接：https://robotics.hellobike.com/papers/2026-07-13-2602-00678
> 频道：papers | 更新日期：2026-07-13 | 标签：arxiv, cs.RO
> 核心摘要：提出结合 MoE 运动策略与 RoboGauge 评估套件的框架，在 Unitree Go2 上验证多地形鲁棒迁移。

# 迈向基于 MoE 的鲁棒四足运动的可靠 Sim-to-Real 可预测性

::: v-pre
> 原文链接：[Toward Reliable Sim-to-Real Predictability for MoE-based Robust Quadrupedal Locomotion](https://arxiv.org/abs/2602.00678)（arXiv 2602.00678 · 发布日期：2026-07-13）

**Toward Reliable Sim-to-Real Predictability for MoE-based Robust Quadrupedal Locomotion**

提出结合 MoE 运动策略与 RoboGauge 评估套件的框架，在 Unitree Go2 上验证多地形鲁棒迁移。

## 核心要点

- 针对复杂地形中的 sim-to-real 差距、奖励过拟合和真机验证风险低效问题。
- MoE 策略用门控专家分解潜在地形与指令建模，仅依赖本体感知实现鲁棒泛化。
- RoboGauge 通过跨地形、难度和域随机化的 sim-to-sim 测试量化迁移性。
- Unitree Go2 在雪地、沙地、楼梯、斜坡和 30 cm 障碍等未见地形上表现鲁棒。

**验证范围与局限**：验证包括 Unitree Go2 真机未见复杂地形与高速测试，RoboGauge 评估主要基于 sim-to-sim 本体感知指标。

## 英文摘要 (Abstract)

Reinforcement learning has shown strong promise for quadrupedal agile locomotion, even with proprioception-only sensing.In practice, however, sim-to-real gap and reward overfitting in complex terrains can produce policies that fail to transfer, while physical validation remains risky and inefficient.To address these challenges, we introduce a unified framework encompassing a Mixture-of-Experts (MoE) locomotion policy for robust multiterrain representation with RoboGauge, a predictive assessment suite that quantifies sim-to-real transferability.The MoE policy employs a gated set of specialist experts to decompose latent terrain and command modeling, achieving superior deployment robustness and generalization via proprioception alone.Robo-Gauge further provides multi-dimensional proprioception-based metrics via sim-to-sim tests over terrains, difficulty levels, and domain randomizations, enabling reliable MoE policy selection without extensive physical trials.Experiments on a Unitree Go2 demonstrate robust locomotion on unseen challenging terrains, including snow, sand, stairs, slopes, and 30 cm obstacles.In dedicated high-speed tests, the robot reaches 4 m/s and exhibits an emergent narrow-width gait associated with improved stability at high velocity.

作者：Tianyang Wu, Hanwei Guo, Yuhang Wang, Junshu Yang, Xinyang Sui, Jiayi Xie, Xingyu Chen, Zeyang Liu, Xuguang Lan
:::

## 原始参考来源 (Sources)

- [arXiv 2602.00678：Toward Reliable Sim-to-Real Predictability for MoE-based Robust Quadrupedal Locomotion](https://arxiv.org/abs/2602.00678)