FOMI force-feedback data interface boosts robot demos 29% over UMI, π0.5 hits 78.4%
Scobleizer · x · 2026-10-26
After a year of robot pretraining, the team released FOMI (Force-feedback Manipulation Interface), a robot-free handheld data collection tool for teaching contact-rich manipulation.
- Inspired by Shuran Song's UMI, it keeps handheld data collection but adds force feedback: operators can feel what the gripper touches — no force sensors, no motors
- Results: 29% more successful demos than UMI in the same time, 18% more than GELLO teleoperation
- π0.5 fine-tuned on 1 hour of FOMI data per task reaches 78.4% average success across four contact-rich tasks
More from Embodied
- Open-source YUBI teleop glove: finger-driven robot gripper with VR tracking, full CAD and BOM released — rsasaki0109 · 2026-10-26
- Asimov 1 humanoid turns one, now shipping to builders in 20+ countries — freelerobot · 2026-10-26
- Yacine has LLM agents auto-generating Blender scenes as RL environments before his quota resets — yacineMTB · 2026-10-26
- Purdue drone tracks runners through dense forest with on-chip LiDAR, no GPS or cloud — chrisfirst · 2026-10-26
- A 16,000 sqft Robot-Native Lab in South Park Opens for Early Robotics Startups — jia_seed · 2026-10-26
- Schaeffler opens 7,500㎡ humanoid robot factory in Taicang, with 5 robots making robot joints — CyberRobooo · 2026-10-26