---
title: "Patch Policy: Efficient Embodied Control via Dense Visual Representations"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.18236v1"
permalink: "https://feed.astralor.com/items/itm_71c8c53bf0caeaa21551a12b"
published: "2026-07-20T17:59:41.000Z"
score: 79
channel: "ai"
tags: ["OpenVLA","Patch Policy","Vision Transformer","Embodied AI","Robot Learning","Attention Mechanism"]
---

# Patch Policy: Efficient Embodied Control via Dense Visual Representations

## AI commentary

> This paper caught my eye: it cleverly enables access to dense visual features at minimal cost, especially friendly for resource-constrained robotics teams. Compared to VLA policies relying on billion-parameter VLMs, Patch Policy's lightweight design could become the preferred paradigm for high-frequency control. If you work on robotic manipulation or visuomotor policies, I strongly recommend reading the full paper.

## At a glance

Patch Policy is a lightweight Transformer policy architecture that directly leverages pretrained dense patch features from Vision Transformers without the need for a full large vision-language model. By using a block-causal attention mask, it preserves temporal causality while allowing the model to attend over multiple patch tokens per observation along with other state information. Across 4 simulated and 3 real-world environment suites, it achieves a 40% relative improvement over global-pooled policies and surpasses fine-tuned OpenVLA-OFT by 18% using only 0.7% of the parameters. This provides the robotics community with a pipeline to efficiently leverage progress in visual representation learning.

## Key points

- Proposes Patch Policy architecture that consumes dense pretrained patch tokens directly, avoiding computational overhead of full VLM.
- Core innovation is a block-causal attention mask balancing temporal causality with dense spatial attention.
- Achieves 40% average improvement across 7 environment suites, surpasses OpenVLA-OFT by 18% with only 0.7% parameters.
- Provides a lightweight, fast pipeline for robotics to leverage visual representation learning progress.

## Full text

Pretrained dense visual features from Vision Transformers (ViTs) are powerful yet have been underutilized in robot learning. Modern robot policies either compress each observation into a single global token, or rely on visual backbones trained from scratch, sacrificing both fine-grained spatial detail and the benefits of large-scale visual pre-training. While there exist policies that do operate on dense patch features like large vision-language-action models (VLAs), they tend to be heavy and slow, inheriting the full cost of a billion-parameter vision-language model (VLM) backbone. We close this gap with Patch Policy, a minimal architectural extension that enables transformer-based policies to consume dense pre-trained patch tokens directly without the computational overhead of a full VLM. At its core is a block-causal attention mask that preserves the temporal causality of standard policies while letting the model attend over many patch tokens per observation, alongside other state information. Patch Policy is lightweight, fast, and highly effective. Across four simulated and three real-world environment suites, our method achieves a 40% relative improvement over policies using state-of-the-art global-pooled representations. Furthermore, it surpasses fine-tuned OpenVLA-OFT by 18% while using roughly 0.7% of the parameters. We believe Patch Policy provides a pipeline for the robotics community to readily leverage continuing progress in visual representation learning, without sacrificing the training efficiency or inference speed required for high-frequency, reactive control. Videos can be viewed at https://patch-policy.github.io

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_71c8c53bf0caeaa21551a12b