The real breakthrough from @asimovinc isn’t just walking. It’s zero-shot sim2real.
A policy trained in simulation can now run directly on Asimov 1 without hours or weeks of hardware-specific tuning.
The interesting part is where the team found the solution: not by forcing the model to compensate for messy hardware, but by making the robot itself more deterministic.
Cleaner CAN communication, stable thermals, accurate actuator models, and better real2sim.
If robots are going to become a serious Physical AI platform, developers need hardware that keeps up with their models.
Asimov is pushing in exactly that direction.
显示更多
