NTU releases ACE-Data-0: 150+ hours of time-synced multimodal embodied data
liuziwei7 · x · 2026-08-29
NTU S-Lab / ACE Robotics released the first part of ACE-Data-0, a large-scale synchronized multimodal dataset now on Hugging Face. It turns real home environments into spatially calibrated, temporally synchronized recording studios, capturing everything on one clock and one world frame:
- Ego (4 fisheye) + 8+ exocentric video views, whole-body and hand motion, object motion, audio, and tactile sensing
- Participants get goal-level instructions ("prepare a cup of tea and serve it"), so planning, hesitation, and improvisation enter the data naturally
- Scale: 150+ hours, 17M+ frames, 75,000+ interaction episodes, 200+ task categories, 50+ participants, 2 environments
- Annotations are metrically tracked/sensed rather than estimated, staying correct under occlusion and motion blur
Blog, tech report (arXiv:2607.28625) released; remaining parts coming soon.
More from Embodied
- Robot Training Fail: Reward Hacking Leads to Distorted Movements — johnowhitaker · 2026-08-29
- Microduck PR absorbs maploc subcrate and dissects why relocalization never worked — victormustar · 2026-08-29
- Microduck BOM Analysis: Is a $10 Price Point Feasible? — NickPassig · 2026-08-29
- HuggingFace shares: Training robot Microduck to perform headstands — huggingface · 2026-08-29
- Magic Chip Design Tutorial: From MOSFET to GDS Generation — Obijuan_cube · 2026-08-29
- AgenticROS Releases Hardware Guide to Build AI-Driven Robots — chrismatthieu · 2026-08-29