Human Demo Data Matches Robot Teleoperation in Embodied AI Training
CyberRobooo · x · 2026-08-06
Beijing Simple AI open-sourced the HiFi-UMI-2K dataset, featuring 2,000 hours of high-fidelity bimanual manipulation data. Built from over 20,000 hours of recording, it contains more than 4.32M processed episodes across 480+ collection scenes, with a 98% valid trajectory rate.
The most surprising finding is that policies trained exclusively on this human demonstration data matched the performance of those trained on real-robot teleoperation data, providing strong evidence for lowering the data collection costs in embodied AI.
Related event: Simple AI Open-Sources 2000-Hour High-Fidelity Robot Dataset(3 posts)→
More from Embodied
- IROS 2026 Human-Robot Dialogue Workshop Extends Submission Deadline to Aug 17 — _jessethomason_ · 2026-08-06
- Colosseum V2: Benchmarking VLA Generalization in Robotic Manipulation — _jessethomason_ · 2026-08-06
- European Student Robotics Association Launches with 2,500 Members Across 8 Countries — MarvinTBaumann · 2026-08-06
- Harness: A Desktop Device to Orchestrate 8 AI Coding Agents — dee_hw · 2026-08-06
- Unitree Seeks Over $900M at $9B Valuation, Leads Global Humanoid Shipments — rohanpaul_ai · 2026-08-06
- Chinese 55kg Humanoid Robot Performs Martial Arts Flips and Side Aerials — rohanpaul_ai · 2026-08-06