Riemann-1.0 Tops Household Robot Leaderboards
量子位 · wechat · 2026-07-19
Riemann Dynamics released Riemann-1.0, a world action model for general-purpose robot manipulation, achieving a new high of 62.6% on RoboCasa-365 to claim SOTA. The article emphasizes that the key isn't simply piling up robot trajectories; instead, it incorporates a massive amount of human egocentric (first-person) video—making up the bulk of its 23.2 million hours of training data—combined with a small amount of real robot and simulation data. This translates "watching how humans work" into executable robotic actions.
The author breaks down the entire engineering pipeline: using an automated pipeline to clean and segment human videos, reconstructing 3D hand trajectories, and finally combining diffusion-based action-video joint modeling with staged training to merge world understanding with embodiment action adaptation. Ablation studies show that introducing human videos boosted the RoboCasa-365 success rate from 48.2% directly to 62.6%; in EgoVLA scenarios, the success rate of long-horizon, multi-stage tasks also jumped from 42.96% to 71.11%.
More from Embodied
- Robot learning researcher joins Mila to work on physical AI and RL — pcastr · 2026-10-27
- Full teardown: inside Zmotor's 32mm linear actuator for humanoids — MarwaEldiwiny · 2026-10-27
- Full Teardown of a Zmotor Linear Actuator Used in China's Humanoid Robots — MarwaEldiwiny · 2026-10-27
- World Model Assigns Weight and Friction to Objects, Then Simulates with a Physics Engine — jonstephens85 · 2026-10-27
- First unboxing shots of Apple's foldable "iPhone Duo" surface — chrisfirst · 2026-10-27
- Bedrock Robotics Raises $600M Series C as Autonomous Excavators Do Paid Work on US Job Sites — davidpattersonx · 2026-10-27