Tag
Briefings and ideas about Simulation-to-real transfer from AI podcast discussions and the research they cite, with links to the original sources.
Josh Gao, CEO of robot-data supplier Mecka AI, says labs train on about a million hours of human data, tens of thousands of gripper hours and a few teleop hours. He argues scaling that alone won't yield a robot that learns any task by watching.
Jason Ma, chief scientist at Dyna Robotics, says robotics has no internet of data. Dyna pretrains on a million hours of human video, adapts each task with minutes to hours of robot data, and Ma expects synthetic environments to add more.
Dyna Robotics' Jason Ma says hour-long robot jobs such as laundry need a planning layer around the low-level action model. He names fragile robot hands, slow hardware testing and vision-only data as limits on more general work.
Keerthana Gopalakrishnan of Google DeepMind said robot training will probably blend teleoperation, sensor-equipped handheld rigs and human video, each trading scale for precision. On world models, she said the jury is still out.
Google DeepMind's Keerthana Gopalakrishnan said humanoid sprint records show locomotion that trains well in simulation. In her view, contact-heavy hand work is the real frontier, where hands have improved but still need reliability and lower cost.
Charlie O'Neill told the Dwarkesh Podcast that AI already learns from deployment, but in slow batches, and that many small updates erase earlier learning; John Schulman expects customers to resist feeding a shared model.
World Labs' Atlas generates video along a user-set camera path and rebuilds scenes from photos, inventing what was not photographed. On Moonshots, Alex inferred from its documentation that it trains on Gaussian splats.