MaxInsights hits 2 million hours of egocentric video to fuel Physical AI scaling
CyberRobooo · x · 2026-09-23
- Physical AI data vendor MaxInsights says it has accumulated 2 million hours of real-world egocentric experience, framing it as a foundation with the next frontier being 10 million hours.
- Its scaling framework has two axes: Experience Scale (volume) and Experience Density (physical information per hour — pose, geometry, object states, contact, force, temporal structure).
- It cites Dyna-2's finding that 1M+ hours of human egocentric video can scale robot learning: performance improved from 1K to 1M hours on both human and robot tasks with no robot data in pretraining.
- Throughput: a distributed pipeline produces up to 450K hours/month across 50,000+ scene types, with structured outputs (environment, task, subtask, instruction).
More from Embodied
- Baidu World 2026 to showcase Apollo Go robotaxi rides and Apollo World physical AI model — Baidu_Inc · 2026-11-12
- NVIDIA's open-source Alpamayo 2 Edge VLA driving model lands in Autoware — drmapavone · 2026-11-12
- A closer look at Astra's robot gripper design (and a joke at its expense) — _Stocko_ · 2026-11-12
- Robot startup passes first presale milestone with 10+ units claimed — R4nsack · 2026-11-12
- Runway launches Praxis-1, a 27B MoE world action model, with a 73ms Fast variant — ChongZzZhang · 2026-11-12
- Dream-1 on bimanual YAM robot folds a towel on voice command at CoRL — DJiafei · 2026-11-12