DiverseUMI-100 open-sourced: 12k tasks from a 1M+ task dataset scaled in 8 weeks

k7agar · x · 2026-10-20

• The team behind a large-scale UMI (Universal Manipulation Interface) data collection effort published the story of scaling from 5 to 90 operators in eight weeks, building a dataset of 1M+ unique tasks and capturing 45k+ new tasks daily.
• They are open-sourcing DiverseUMI-100: 12,000+ tasks from a 100-hour sample, complete with dense annotations from Argus, their robotics annotation pipeline.
• Key argument: robotic world models require far higher task diversity than imitation learning, which motivates the massive collection operation. A long thread explains the why and how.

Original post →

More from Embodied

• KAIST unveils next-gen humanoid: 12km/h walking, moonwalk, eyes-closed balance — CyberRobooo · 2026-10-20
• AnyGoal: benchmarking world model generalization beyond expert rollouts — galilai-group · 2026-10-20
• EgoXperience: Single-Pass Joint Reconstruction of Camera, Scene, Hands and Objects from Egocentric Video — ropedia-ai · 2026-10-20
• KAIST Humanoid v0.9 pushes 35kg objects while walking using onboard visual feedback — CyberRobooo · 2026-10-20
• 15 Robotics Research Accounts Worth Following on X, From VLA Datasets to Teleop — FinanceYF5 · 2026-10-20
• DIY 'ESP32 Roadster': a mini Tesla Roadster built on an ESP32 — glenbeer · 2026-10-20

Embodied channel →