ACE-Data-0 hits 120K downloads in first month: 17M-frame multimodal embodied AI dataset
liuziwei7 · x · 2026-09-18
ACE-Data-0, an L5 multimodal Ego-Exo dataset for embodied AI from ACE Robotics (S-Lab, Nanyang Technological University), has surpassed nearly 120K downloads on Hugging Face in its first month.
Key facts:
- 150+ hours recorded in real home environments: 17M+ video frames, 75K+ interaction segments, 200+ task categories, 50+ participants
- Each moment is captured from 8+ exocentric views plus 4 egocentric fisheye cameras
- Modalities span egocentric and third-person vision, human motion, 6DoF object trajectories, audio, and tactile signals
Methodologically, ACE turns real homes into spatially calibrated, time-synchronized recording studios. Participants get goal-level instructions (e.g., "prepare a cup of tea and serve it at the table") instead of step-by-step scripts, so planning, hesitation, and improvisation enter the data naturally. Human states, object states, and contact are metrically tracked or directly sensed rather than estimated from images, keeping annotations correct under occlusion, extreme viewpoints, and motion blur.
The dataset is positioned as an open foundation for imitation learning, world models, VLA, and robot policy learning, under a research license.
More from Embodied
- Stanford's FAMOS infers 3D articulation from sparse partial point clouds feed-forward — Stanford-University · 2026-09-18
- Waymo to launch autonomous robotaxis in Singapore by 2028, its first Southeast Asia market — emmanuelvivier · 2026-09-18
- Agility Robotics unveils Digit 5 humanoid built to safely work in contact with humans — emmanuelvivier · 2026-09-18
- Self-driving project spent 87.5% of time in simulation; imitation learning beat lane estimation — abursuc · 2026-09-18
- CARLA driving expert with privileged maps trains vision-based imitation policy — abursuc · 2026-09-18
- AMP reframes robot manipulation as pixel classification to dodge action-space explosion — jiqizhixin · 2026-09-18