# 估计，不要模仿：复用可微分基于状态的策略用于视觉运动控制

> 原文链接：https://robotics.hellobike.com/papers/2026-09-27-2609-34018
> 频道：papers | 更新日期：2026-09-27 | 标签：arxiv, cs.RO
> 核心摘要：本文通过训练带动作一致性损失的视觉状态估计器复用状态专家，并在五个操作任务和真实Panda上验证。

# 估计，不要模仿：复用可微分基于状态的策略用于视觉运动控制

::: v-pre
> 原文链接：[Estimate, Don't Imitate: Reusing Differentiable State-Based Policies for Visuomotor Control](https://arxiv.org/abs/2609.34018)（arXiv 2609.34018 · 发布日期：2026-09-27）

**Estimate, Don't Imitate: Reusing Differentiable State-Based Policies for Visuomotor Control**

本文通过训练带动作一致性损失的视觉状态估计器复用状态专家，并在五个操作任务和真实Panda上验证。

## 核心要点

- 仿真训练策略依赖特权状态信息，部署时只能利用噪声相机图像等部分观测。
- 方法冻结可微状态专家，学习视觉状态估计器，并结合状态监督与动作一致性损失。
- 调度目标先建立有物理意义的状态估计，再逐步强调会影响专家动作的误差。
- 在五个目标条件操作任务中，复用专家持续优于直接像素到动作模仿；真实Panda上实现sim-to-real迁移。

**验证范围与局限**：验证范围包括五个目标条件操作任务，以及一台真实Panda机器人上的sim-to-real迁移。

## 英文摘要 (Abstract)

Simulation-trained manipulation policies can exploit privileged state information to learn effective contact-rich behaviours, but deployment requires acting from partial observations such as noisy camera images. A common solution is teacher-student distillation, in which a visuomotor policy is trained to reproduce the actions of the privileged expert. This requires the student to jointly infer the task-relevant state and relearn the expert's action mapping that is already available. An alternative is to reuse the state-based expert and learn only a perceptual interface that reconstructs its missing state inputs. However, minimising the state estimate error alone does not necessarily minimise the downstream control error induced by these estimates. To bridge this gap, we train a visual state estimator using both direct state supervision and an action-consistency loss backpropagated through the frozen, differentiable expert. A scheduled objective first establishes a physically meaningful state estimate and progressively emphasises errors that affect the expert's actions. Across five goal-conditioned manipulation tasks, retaining the expert consistently outperforms direct pixel-to-action imitation from the same expert demonstration corpus. We further demonstrate sim-to-real transfer on a physical Panda robot, achieving 76% success without retraining the underlying expert.

作者：Denis Shcherba, Adrian Abel, Eckart Cobo-Briesewitz, Paul Mattes, Wojciech Samek, Marc Toussaint
:::

## 原始参考来源 (Sources)

- [arXiv 2609.34018：Estimate, Don't Imitate: Reusing Differentiable State-Based Policies for Visuomotor Control](https://arxiv.org/abs/2609.34018)