UMI step 3: video+IMU SLAM recovers gripper trajectories under 1cm error
gabriberton · x · 2026-11-03
Step 3 of the UMI pipeline: compute the grippers' trajectories during the task. IMU alone is too noisy, so use a SLAM method fusing video and IMU. The MoCap benchmark figure shows mean SLAM tracking errors of 6.1mm/3.5° per-gripper ATE and 10.1mm/0.8° inter-gripper RPE — under 1 cm and 4° overall.
Related event: UMI: The Data-Collection Foundation Behind Many Robot Startups(6 posts)→
More from Embodied
- Apps still not optimized for iPhone Duo's dual screen weeks after launch — evielync · 2026-11-03
- UMI: the data collection paper many robotics startups are built on — gabriberton · 2026-11-03
- New startup Shiqu Tech unveils android robot head "Unit 01" — TinfoilTricorn · 2026-11-03
- Demo app for 'vibe designing' robot clothing simulates cloth physics, thermodynamics and gaits — kscottz · 2026-11-03
- Yacine MTB says baby nuked his time but a Sunday off yielded 'ridiculous progress' on his drone hobby project — yacineMTB · 2026-11-03
- Would you fight a robot? Turing Award winner David Patterson joins the joke — davidpattersonx · 2026-11-03