NVIDIA’s GTC 2025 Announcement for Physical AI Developers: New Open Models and Datasets
What Happened
NVIDIA’s GTC 2025 Announcement for Physical AI Developers: New Open Models and Datasets
Fordel's Take
NVIDIA released updated Isaac GR00T model weights and new synthetic data pipelines for physical AI at GTC 2025, targeting robotics manipulation and locomotion tasks.
For teams building sim-to-real transfer in Isaac Sim, pretrained GR00T checkpoints cut baseline training cost significantly — starting from scratch on a manipulation policy is now avoidable. This is irrelevant to the 95% of developers building RAG, agents, or inference APIs.
Robotics teams working on manipulation or locomotion with NVIDIA Isaac should pull the updated checkpoints. Everyone else can skip this cycle.
What To Do
Pull updated GR00T checkpoints for your Isaac simulation pipeline instead of training a manipulation policy from scratch because the pretrained weights substantially reduce sim-to-real transfer time.
Cited By
React
Get the weekly AI digest
The stories that matter, with a builder's perspective. Every Thursday.