A new Physical AI player: FSD-grade team unveils first model Simate-beta, lands on RoboDojo

A new player has entered the Physical AI race. According to a report by QbitAI, a team described as having FSD-level experience has unveiled its first Physical AI model, Simate-beta, and dropped it straight into the RoboDojo platform.

The more interesting part is the engineering stack behind it. Simate plugs the whole loop — training, inference and evaluation — into its own in-house infrastructure, giving the team a closed pipeline running from data to assessment. For a field that depends on repeated experimentation, that completeness often decides whether iteration can be sustained.

That machinery shows up in scheduling. The source says the system uses extremely tight task orchestration and resource scheduling to run dozens of independent research lines in parallel, turning model exploration into a large-scale parallel search rather than a serial one.

Why it matters: Physical AI pushes model capability out of the screen and into the physical world, where data loops, simulation, evaluation standards and compute scheduling all matter far more than in pure text work. A team with FSD-grade background tends to bring end-to-end modelling, simulation and large-scale data engineering habits along with it.

RoboDojo matters here too. The source frames the launch as the model arriving together with its platform, which suggests model and evaluation ground are being advanced as one package. For Physical AI, reusable training-and-evaluation infrastructure often determines iteration speed more than any single model does.

Simate-beta is still a first release, and the source gives no parameter count, benchmark score or open-source plan. Three things to watch: whether a stable version follows quickly, whether RoboDojo publishes comparable evaluation results, and whether the in-house stack is opened up to outsiders.