Memories.ai turns first-person video into Physical AI training data, 3 of top 5 robotics labs on board
CyberRobooo · x · 2026-11-11
Memories.ai launched MemoriesUpload, converting internet-scale first-person video of humans doing tasks into searchable Physical AI training data that tracks each hand's actions to improve humanoid dexterity. The founder says collecting such data at quality requires hardware, models, and infra built together—over a year of work—and three of the top five frontier robotics labs are already training on the data. The thesis: curated egocentric video at scale could set the performance ceiling for humanoid robots.
More from Embodied
- OmniForge concept: one machine to 3D print, weld, solder, CNC and sew — flowersslop · 2026-11-11
- Humanoid rides a skateboard with no on-board sensors, stunning viewers — jBby_omg · 2026-11-11
- RoboScript hits 83.4% success on RoboTwin 2.0 with one demo per task, beating trained policies by 20+ points — H-EmbodVis · 2026-11-11
- DynaRobotics' Taku Robot Performs Zero-Shot Long-Horizon Mobile Manipulation On-Site — JasonMa2020 · 2026-11-11
- EgoForce dataset: 2,000 hours of egocentric video paired with calibrated tactile forces — yining_hong · 2026-11-11
- Robot Dyna spins the laundry dial and presses start in new demo — JasonMa2020 · 2026-11-11