HoMMI: Learning Whole-Body Mobile Manipulation
leto__jean · x · 2026-07-16
This post introduces the team's **HoMMI (Whole-Body Mobile Manipulation Interface)**, which aims to learn whole-body mobile manipulation directly from human demonstrations. Key highlights include: - Combines **egocentric + UMI** data. - Requires zero teleoperation (0 teleop), learning **dual-arm + whole-body manipulation** capabilities directly from human demos. - Covers **long-horizon navigation** and **active perception**. - This work will be presented at relevant RSS sessions and the poster session.
More from Embodied
- HarmoHOI generates multi-view hand-object videos and aligned 3D motion in one diffusion model — cn-scut · 2026-07-21
- Tesla is reportedly building a humanoid robot factory aimed at 10 million units a year — davidpattersonx · 2026-07-21
- Anthropic is reportedly in talks to buy Physical Intelligence — MarvinTBaumann · 2026-07-21
- IROS 2026 workshop sets Aug. 10 paper deadline for embodied world models and WorldArena 2.0 — 量子位 · 2026-07-21
- Work Louder’s new input device draws praise for its multi-input design — Dimillian · 2026-07-21
- Sudo Tech shows 10+ embodied AI skills at WAIC and ties them to a real CATL line — 机器之心 · 2026-07-21