# E2DRL实现四足机器人端到端3D避障并完成Go2真机验证

> 原文链接：https://robotics.hellobike.com/papers/2026-10-15-2610-19627
> 频道：papers | 更新日期：2026-10-15 | 标签：auto, arxiv
> 核心摘要：提出E2DRL框架解决腿足机器人视觉避障梯度断裂难题，通过软可微碰撞与SAPO教师策略蒸馏，在Unitree Go2上首次实现端到端可微强化学习的3D避障Sim-to-Real部署。

# E2DRL实现四足机器人端到端3D避障并完成Go2真机验证

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots](https://arxiv.org/abs/2610.19627)（arXiv 2610.19627）

## 核心内容

针对腿足机器人端到端视觉避障中硬碰撞约束导致梯度失效的问题，研究提出E2DRL框架。该方法采用软可微障碍物交互公式，结合速度条件停止梯度整形提供一阶训练信号。实验表明，E2DRL样本效率优于PPO，3D避障性能超越PPO及基于对偶的DCBF NMPC算法，并首次在Unitree Go2真机上完成静态与动态环境下的Sim-to-Real验证。

系统架构包含特权教师与深度感知学生网络：教师策略经Soft Analytic Policy Optimization (SAPO) 在可微仿真器中训练，再通过DAgger蒸馏为循环深度学生网络，最后利用地形感知门控模块与基础运动策略融合部署。该设计有效解决了真实环境中深度感知噪声与控制延迟问题，确保了从仿真到Unitree Go2实体机器人的策略迁移稳定性与3D空间避障能力。

## 关键看点

- **软可微碰撞**：引入多点穿透惩罚与速度条件停止梯度整形，解决硬碰撞导致的梯度断裂与局部极小值陷阱
- **策略蒸馏架构**：采用SAPO训练特权教师策略，经DAgger蒸馏为循环深度学生网络，适配真实传感器输入
- **真机验证突破**：在Unitree Go2上首次实现端到端可微RL的3D避障Sim-to-Real，覆盖静动态场景
- **性能基准对比**：样本效率优于PPO，3D避障与速度跟踪性能显著超越PPO及DCBF NMPC传统方法

## 工程与产业观察

E2DRL通过软可微化重构碰撞约束，打通了端到端视觉避障的梯度传播链路，并在Go2上验证了Sim-to-Real可行性。其“SAPO教师+DAgger学生+地形门控”架构为具身智能提供了兼顾训练效率与部署鲁棒性的新范式，对复杂非结构化环境下的腿足导航具有直接工程参考价值。

## 英文摘要 (Abstract)

Differentiable reinforcement learning shows strong promise for robot policy learning, but its application to legged locomotion has remained limited to blind walking or simulated 2D tasks. Extending it to end-to-end visual obstacle avoidance is challenging because hard collision constraints and early termination upon contact corrupt first-order gradients, trapping optimizers in local minima. We present E2DRL, which addresses this with a soft differentiable obstacle-interaction formulation: penetration penalties across multiple body keypoints with velocity-conditioned stop-gradient shaping to provide the first-order training signals for 3D obstacle avoidance. To realize a deployable perception-control system, a privileged teacher trained via Soft Analytic Policy Optimization (SAPO) in a differentiable simulator is distilled into a recurrent depth-based student via DAgger, then blended with a base locomotion policy using a terrain-aware gating module for deployment. In simulation, E2DRL achieves higher sample efficiency than PPO and demonstrates superior velocity-tracking performance and 3D obstacle-avoidance capability compared with both PPO and Duality-based DCBF NMPC. We present the first sim-to-real validation of end-to-end differentiable RL for 3D obstacle avoidance on a Unitree Go2 in static and dynamic environments.

作者：Cheng Fan, Bingyang Zhou, Gang Yang, Zixuan Zhuang, Aoqian Zhang, Pengzhi Yang, Xin Liu, Chunzheng Wang, Yan Rui Tan, Fan Shi
:::

## 原始参考来源 (Sources)

- [arXiv 2610.19627：E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots](https://arxiv.org/abs/2610.19627)