The promise of physical AI is tantalising: programming robots as seamlessly as software. Yet, robots still struggle with the messy unpredictability of the real world.

Antioch, a startup that builds simulation tools for robot developers, hopes to close this 'sim-to-real' gap. Their seed round of $8.5 million aims to provide realistic virtual environments where robots can be trained effectively.

The challenge lies in making these simulations as accurate as the real world. Antioch uses models from Nvidia and World Labs, adapting them for specific domains like sensor and perception systems. This approach is crucial given the high stakes of physical AI: safety cases and high-accuracy tasks require flawless simulations.

Antioch’s cofounders bring experience from major tech companies like Meta Reality Labs and Google DeepMind, positioning them well to deliver on this promise. With backing from major investors, Antioch aims to democratise access to robust simulation tools for startups and large multinationals alike.