How Does Dyna Robotics Generate Synthetic Robot Data?

Jason Ma says his team generates synthetic robot data with frontier models that build synthetic environments and with a world action model that makes its own training videos. Synthetic environments, synthetic data, and simulation data are one of the next frontiers for robotics. Egocentric data and robot data are already scaling, and synthetic data is a new dimension of scaling.

In his PhD he focused on simulation-based training for robotics, and simulation was not ready. Eureka was the first paper to take a large language model, GPT-4, to self-generate environments and reward functions for reinforcement learning in simulation. The demo was a robot spinning pencils. The model lacked good vision and an innate 3D understanding, so it generated code from limited text feedback. Frontier model vision is now strong enough for real-to-sim, CADing, and 3D reconstruction. The team treats the shift as a way to get tailored data while building models and hardware, and the technology is worth going very deep into.

They have also used their own world action model. The model generates actions and videos, so it can generate its own training data, a self-fulfilling flywheel. They have scaled to a million hours of data and are deploying robots that already do cool things, still only the tip of the iceberg.