Generative AI will help robots perform actions based on input — a combination of text and speech, video, «human demonstrations» and previous experience.
As noted by ReutersThe generative AI components, called Project GR00T, will complement Nvidia’s existing hardware and software platform for robotics.
«Smarter, faster robots will power heavy industry around the world,» says Rev Lebaredian, vice president of Omniverse and Simulation Technology. «We are collaborating with the entire global robot ecosystem to accelerate development and deployment».
«Jetson Thor» — is the powerful computer behind the software of this generative artificial intelligence and is based on one of Nvidia’s AI chips. According to the company, it will provide robots with enough computing power to perform complex tasks and interact with humans and machines.
The tools include the ability to teach robots to make better decisions over time — a process known as «reinforcement learning». Nvidia also said it will release pre-trained robotic device models and other software that will improve robotic arm functions and multi-camera sensing capabilities.
A week ago, Figure AI startup demonstrated the first results of cooperation with OpenAI — in the video, the robot Figure 01 easily perceived requests by sight and hearing, and provided full answers or performed actions in response to human requests.