From NVIDIA: 2024-11-06 11:00:44

NVIDIA unveiled new AI and simulation tools at CoRL in Munich to accelerate robot development. The NVIDIA Isaac Lab framework, Project GR00T for humanoid robots, and world-model development tools were showcased. The Cosmos tokenizer provides superior visual tokenization, while NeMo Curator offers faster video processing. NVIDIA also presented 23 papers and workshops at the conference.

Isaac Lab, an open-source robot learning framework, is built on NVIDIA Omniverse for training robot policies at scale. Leading robot makers and research entities are adopting Isaac Lab for various embodiments. Project GR00T aims to accelerate humanoid robot development with six new workflows focusing on capabilities like dexterity and mobility. NVIDIA also released new development tools for world model builders.

NVIDIA Cosmos tokenizers provide efficient encoding for building world models, enabling high-quality video and image reconstructions. 1X Technologies updated their dataset using the Cosmos tokenizer for efficient training. NeMo Curator now includes a video processing pipeline to improve world-model accuracy. The tools are being used by humanoid and general-purpose robot developers.

NVIDIA released nearly two dozen research papers at CoRL covering breakthroughs in robot learning. Papers include integrating vision language models and developing long-horizon planning strategies. SkillGen and HOVER are notable projects for training robots and controlling humanoid robot locomotion. NVIDIA researchers will participate in nine workshops at the conference.

NVIDIA Isaac Lab 1.2 and the Cosmos tokenizer are available now, with NeMo Curator for video processing coming soon. The new Project GR00T workflows will be released shortly to assist robot companies. Developers can access guides and tutorials for Isaac Lab and apply to join the NVIDIA Humanoid Robot Developer Program.



Read more at NVIDIA: NVIDIA Advances Robot Learning, Humanoid Development With New AI and Simulation Tools