Nvidia’s Cosmos-Transfer1 makes robot training freakishly realistic—and that changes everything
1 min read
Summary
Nvidia has launched an AI model that will help developers create simulations for training autonomous vehicles and robots.
Cosmos-Transfer1 allows varying weights to be put on different aspects of an AI-simulated scene, meaning a balance can be struck between allowing a robot arm to move naturally, whilst having a more diverse background environment.
The model can be used to generate real-time simulations, another powerful tool for testing AI in a variety of scenarios.
Nvidia has made the model – and its underlying code – freely available on GitHub, in keeping with its policy of supporting open-source innovation.
It is the latest component of Nvidia’s Cosmos platform, a suite of AI tools aimed at physical AI development.