Nvidia unveiled its new robot model on Monday at the Conference on Robot Learning in Austin, Texas, introducing an open system trained partly on 100,000 hours of human video to teach machines delicate hand skills, according to The Next Web.
Key takeaways
- Nvidia introduced the GR00T 2 open AI model for robot hand skills.
- The system was trained partly on 100,000 hours of first-person human video.
- Foxconn, Siemens, and Nexuni are currently testing the new software.
- Nvidia also released Isaac Lab 3.0 for simulation training.
What is the new robot model from Nvidia?
The Next Web reports that Nvidia calls GR00T 2 an open world action model because it predicts how its actions will change what it sees before it moves. The company trained the software on robot data and first-person footage, which Nvidia is releasing for public use. This lets machines learn tasks from people and share those skills with robots that have different physical bodies.
How well does the Nvidia GR00T 2 system perform tasks?
During tests conducted by Nvidia, the software completed 40 of 50 tasks at least once without any prior training for those specific tasks. When shown a task a single time, it reached an average success rate of 71 percent. That number rose to 95 percent on average when machines watched five demonstrations, though Nvidia noted those tasks were selected from ones the model could already learn effectively.
What happens next for the technology?
Companies are already putting the software through trials for real-world applications. Foxconn is using the technology to help build AI servers, Siemens is testing it for factory work, and Nexuni is trying it out for laundry tasks. Meanwhile, Ken Goldberg, a robotics professor at UC Berkeley who collaborated with Nvidia on two projects, called agentic robotics the most exciting technological development since the internet in 1995.
Frequently asked questions
Nvidia GR00T 2 is an open AI model that helps robots learn delicate hand skills by predicting how actions change what the robot sees.
Nvidia announced the new model at the Conference on Robot Learning held in Austin, Texas.
The system was trained partly on 100,000 hours of first-person human video footage released by Nvidia.
Foxconn is using it to help build AI servers, Siemens is testing it for factory work, and Nexuni is trying it for laundry.
Isaac Lab 3.0 is an open-source tool released by Nvidia for training robots in simulation up to 2.5 times faster.
Based on reporting by The Next Web. Read the original article
