Maxinsights report: Physical AI data scaling laws — after 1M hours, it's experience density
rohanpaul_ai · x · 2026-09-22
Maxinsights' research report "Beyond 1M Hours: The Data Scaling Laws of Physical AI" argues data, not models or compute, is the defining bottleneck for robotics — high-quality robot demos are expensive and embodiment-bound.
Key points:
- Dyna-2 (Dyna Robotics), pretrained on 1M+ hours of human egocentric video, improves monotonically across four orders of magnitude and survives the embodiment gap; Genesis AI's GENE-26.5 breaks it from another direction.
- The next frontier scales two coupled axes: Experience Scale (hours recorded) and Experience Density (learnable physical information per hour — object states, contact, forces, geometry, timing, tool use).
- Equal-duration recordings can differ an order of magnitude in learning value; at 10M hours that's the difference between a corpus and an archive.
More from Embodied
- UBTECH sells 400 U1 Pro humanoids to Yongda Auto in ~$10M deal — CyberRobooo · 2026-11-12
- 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