ActiveGlasses: Learning Manipulation with Active Vision from Ego-centric Human Demonstration

TL;DR

ActiveGlasses introduces a system using ego-centric vision via smart glasses for natural human demonstration and zero-shot robot manipulation transfer, addressing scalability and embodiment challenges in robot learning.

Contribution

It presents a novel active vision system with a single stereo camera on glasses for data collection and policy transfer, enabling natural demonstrations and cross-platform robot manipulation.

Findings

Achieves zero-shot transfer with active vision in complex tasks.

Outperforms strong baselines under the same hardware setup.

Generalizes across two robot platforms.

Abstract

Large-scale real-world robot data collection is a prerequisite for bringing robots into everyday deployment. However, existing pipelines often rely on specialized handheld devices to bridge the embodiment gap, which not only increases operator burden and limits scalability, but also makes it difficult to capture the naturally coordinated perception-manipulation behaviors of human daily interaction. This challenge calls for a more natural system that can faithfully capture human manipulation and perception behaviors while enabling zero-shot transfer to robotic platforms. We introduce ActiveGlasses, a system for learning robot manipulation from ego-centric human demonstrations with active vision. A stereo camera mounted on smart glasses serves as the sole perception device for both data collection and policy inference: the operator wears it during bare-hand demonstrations, and the same camera is mounted on a 6-DoF perception arm during deployment to reproduce human active vision. To enable zero-transfer, we extract object trajectories from demonstrations and use an object-centric point-cloud policy to jointly predict manipulation and head movement. Across several challenging tasks involving occlusion and precise interaction, ActiveGlasses achieves zero-shot transfer with active vision, consistently outperforms strong baselines under the same hardware setup, and generalizes across two robot platforms.