@tlensnap: NVIDIA has introduced GR00T 2, an advanced AI model designed to help robots imagine what will happen in the physical world before taking action. Instead of teaching robots every task from scratch, GR00T 2 uses a World-Action Model (WAM) to predict how the environment will change and generate the precise movements needed to complete tasks. What makes this technology impressive is its ability to work across different robotic bodies, from simple parallel grippers to advanced 22-degree-of-freedom hands and full humanoid robots. Trained on more than 100,000 hours of first-person human video, it allows robots to learn skills by watching humans perform everyday tasks. Its performance is equally impressive. The system achieves 40% success on 50 completely new tasks without prior demonstrations, rising to 95% after just five demonstrations and a short fine-tuning session. It can even perform delicate operations, including installing GB300 server memory modules, striking matches, and assembling multi-part toys. With AI-powered visual guidance, its step-by-step accuracy can also improve from 40% to 92%. This could bring robots one step closer to learning, adapting, and performing complex tasks in the real world. Source: NVIDIA #nvidia #robotics #physicalai #humanoidrobots #artificialintelligence

Tlensnap

Open In TikTok:

Region: PK

Thursday 12 November 2026 12:50:37 GMT