EgoGuide: Egocentric Guidance for Efficient Robot-Free Demonstration Collection and Learning

2026-06-12

AI Summary

This work addresses the inefficiency, high redundancy, and lack of global context in data collection when robots learn from real-world demonstrations. To overcome these limitations, the authors propose a synchronized multi-view data acquisition framework that integrates wrist-mounted and head-mounted (egocentric) cameras, complemented by an online visual-geometric data quality assessment mechanism to enhance collection efficiency. Furthermore, they introduce a gated egocentric residual policy network that effectively fuses global scene information to correct observations degraded by local ambiguities or occlusions. The proposed approach significantly reduces the number of required demonstrations, improves data efficiency, and enhances policy robustness under visual occlusion conditions.

Abstract

Robot learning from real-world demonstrations is currently constrained by data scaling. Universal Manipulation Interface (UMI) provides an efficient robot-free data collection interface, yet current UMI-style pipelines often collect redundant demonstrations and lack global scene context. To improve data efficiency, we present EgoGuide, a collection interface that records synchronized wrist and head/egocentric observations and couples them with online visual-geometric data quality guidance. We also introduce a Gated Egocentric Residual Policy for robust learning from a viewpoint-varying egocentric camera, allowing head/egocentric context to correct ambiguous local observations while preserving stable wrist-view control. Real-world experiments show that EgoGuide reduces the required number of data episodes and improves data efficiency. The residual policy further improves robustness under visual occlusion. Project Page: https://silicx.github.io/EgoGuide

Problem

• robot learning
• demonstration collection
• data efficiency
• egocentric vision
• scene context

Innovation

• Egocentric Guidance
• Robot-Free Demonstration
• Data Efficiency
• Residual Policy
• Visual-Geometric Guidance