# 用基于采样的模型预测控制加速视觉策略学习

> 原文链接：https://robotics.hellobike.com/papers/2026-09-17-2609-20575
> 频道：papers | 更新日期：2026-09-17 | 标签：arxiv, cs.RO
> 核心摘要：针对视觉运动与操作策略学习开销大且易陷入非预期接触的问题，SGPS结合采样式MPC与一阶策略优化，并在仿真和真实Go2上验证。

# 用基于采样的模型预测控制加速视觉策略学习

::: v-pre
> 原文链接：[Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control](https://arxiv.org/abs/2609.20575)（arXiv 2609.20575 · 发布日期：2026-09-17）

**Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control**

针对视觉运动与操作策略学习开销大且易陷入非预期接触的问题，SGPS结合采样式MPC与一阶策略优化，并在仿真和真实Go2上验证。

## 核心要点

- 视觉运动和操作需协调环境接触，训练计算与显存成本较高。
- SGPS用采样式MPC反复细化动作目标，并结合短时域FoPG更新。
- 解耦FoPG将渲染排除出计算图，可直接从深度观测学习。
- 在仿真Go2/G1多任务验证，细化提升学习；蒸馏策略零样本迁移真机。

**验证范围与局限**：验证范围包括仿真Unitree Go2和G1的运动、越障、推箱和双臂搬运，以及真实Go2的自主小跑、爬行、跨越障碍和行为切换。

## 英文摘要 (Abstract)

Learning visual policies for locomotion and manipulation requires coordinating contact with the environment and can incur substantial computation and GPU memory costs. First-order policy gradients (FoPG) reduce training cost through differentiable simulation, but local optimization can converge to unintended contact patterns. To address this shortfall, we propose Sampling-Guided Policy Search (SGPS), which couples recurring action-target refinement by sampling-based model-predictive control with first-order policy optimization. Behavior cloning initializes the policy from sampled actions; training then alternates sampling-based refinement with short-horizon FoPG updates under perturbed initial states and randomized dynamics. For visual policy training, we use a decoupled FoPG formulation that excludes rendering from the computation graph, enabling direct learning from depth observations without a state-policy teacher. On a single GPU, SGPS learns policies for locomotion, obstacle traversal, crate pushing, and bimanual carrying on simulated Unitree Go2 and G1 robots. Our experiments further show that refinement improves policy learning beyond initialization and tracking alone. For hardware deployment, the distilled policy transfers zero-shot to a real Go2 and uses onboard depth to autonomously trot, crawl, clear hurdles, and switch between these behaviors.

作者：Yilang Liu, Haoxiang You, Qian Wang, Daniel Rakita, Ian Abraham
:::

## 原始参考来源 (Sources)

- [arXiv 2609.20575：Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control](https://arxiv.org/abs/2609.20575)