---
title: "Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.25895v1"
permalink: "https://feed.astralor.com/items/itm_4d813e350039d2ae6ba505c3"
published: "2026-07-28T15:52:02.000Z"
score: 83
channel: "ai"
tags: ["HiFi-UMI","SLAM","UMI","Robotic Manipulation","Policy Learning"]
---

# Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone

## AI commentary

> The most striking point of this paper is proving that pure robot-free data can match real teleoperation performance, breaking the convention of needing a small real-robot anchor for post-training. For research groups looking to reduce hardware costs and industrial robot application developers, HiFi-UMI's data production system and open-source dataset are directly usable resources. Pay attention to the engineering details of head-mounted SLAM and microsecond synchronization—they are key to achieving the reported accuracy.

## At a glance

Deployable manipulation policies are bottlenecked by data that is both high-fidelity and scalable. Real-robot teleoperation is accurate but costly; robot-free UMI capture is scalable but imprecise. This paper presents HiFi-UMI, a portable data-production system achieving 3 mm workspace-local accuracy via head-mounted offline stereo-inertial SLAM, native relative pose, microsecond GPIO triggers, and 200-degree-per-hand wide-angle cameras. Using this data, policies post-trained solely on HiFi-UMI demonstrations deploy directly on real robots, matching teleoperation baselines across three backbones (success rate differences from -2.5 to +3.1 pp), with the best achieving 85% on a precision insertion task. Pre-training on 4,000 hours reduces action error on ten unseen tasks by 41%. The open-source HiFi-UMI-2K dataset provides 2,000 hours of microsecond-synchronized, ultra-wide-FoV demonstrations.

## Key points

- HiFi-UMI achieves 3 mm end-effector accuracy via head-mounted SLAM, native relative pose, microsecond triggers, and wide-angle cameras without external tracking.
- Policies post-trained solely on HiFi-UMI data deploy directly on real robots, matching teleoperation baselines across three backbones.
- Pre-training on 4,000 hours of the same corpus reduces action error by 41% on ten unseen tasks and boosts real-robot success by 18.1 pp on StarVLA-QwenPI.
- The open-source HiFi-UMI-2K dataset provides 2,000 hours of microsecond-synchronized, ultra-wide-FoV demonstrations, automatically reconstructed and validated via simulation replay.

## Full text

Learning deployable manipulation policies is bottlenecked by the scarcity of data that is both high-fidelity and scalable. Real-robot teleoperation is accurate but costly to scale; robot-free UMI capture scales readily, and current practice uses the resulting data mainly for pre-training, adding a small real-robot "anchor" at post-training. We ask whether raising the fidelity of robot-free UMI data, rather than shrinking the real-robot fraction, can remove that anchor. We present HiFi-UMI, a portable UMI data-production system co-designed for trajectory accuracy, inter-gripper relative pose, synchronization, and field of view: head-mounted offline stereo-inertial SLAM, native rather than reconstructed relative pose, a shared microsecond GPIO trigger, and two wide-angle cameras per hand covering ~200 degrees. It reaches 3 mm workspace-local end-effector accuracy without external tracking infrastructure. Using this corpus, we demonstrate zero-robot post-training: a policy post-trained solely on HiFi-UMI demonstrations deploys directly on a real robot and matches in-domain teleoperation across three backbones spanning the vision-language-action and world-action-model families, with success-rate differences of -2.5, +3.1, and -0.6 percentage points on StarVLA-QwenPI, OpenPI-pi_0.5, and LingBot-VA; the strongest policy reaches 85% on a precision insertion task, even though the teleoperation baseline is collected in the evaluation scene and no HiFi-UMI trajectory is. Pre-training on 4,000 hours from the same corpus lowers action error on ten unseen tasks by 41% and, on StarVLA-QwenPI, raises real-robot success by a further 18.1 percentage points. We open-source HiFi-UMI-2K, 2,000 hours of microsecond-synchronized, ultra-wide-FoV demonstrations, each automatically reconstructed and validated through simulation replay, as a large-scale, high-fidelity resource for the robot-learning community.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_4d813e350039d2ae6ba505c3