Pixel-UMI: Learning Environment-Aware Whole-Body Manipulation from Sensorized Human Demonstrations
arXiv cs.AI · 3 Nov 2026, 04:59 AEST
Pixel-UMI is a new pipeline designed for learning environment-aware whole-body manipulation from human demonstrations. It addresses challenges in policy learning due to embodiment differences and aims to improve robot performance in various manipulation tasks.
Key facts
- Sensorized human demonstrations offer scalable data for mobile manipulation, but embodiment differences complicate policy learning.
- We present Pixel-UMI, a pipeline for learning environment-aware whole-body policies from human data.
- We evaluate Pixel-UMI on five challenging tasks spanning articulated-object manipulation, deformable-object handling, constrained-space reaching, whole-body collision avoidance, and contact-rich bimanual coordination, achieving 87.0% average task progress.
Read aloud
Pixel-UMI is a new pipeline designed for learning environment-aware whole-body manipulation from human demonstrations. It addresses challenges in policy learning due to embodiment differences and aims to improve robot performance in various manipulation tasks. Sensorized human demonstrations offer scalable data for mobile manipulation, but embodiment differences complicate policy learning. We present Pixel-UMI, a pipeline for learning environment-aware whole-body policies from human data. We evaluate Pixel-UMI on five challenging tasks spanning articulated-object manipulation, deformable-object handling, constrained-space reaching, whole-body collision avoidance, and contact-rich bimanual coordination, achieving eighty seven point zero percent average task progress.
Sources in this cluster
- Pixel-UMI: Learning Environment-Aware Whole-Body Manipulation from Sensorized Human Demonstrations
arXiv cs.AI · 3 Nov 2026, 04:59 AEST