Louisa Crawford
6 November 2024 16:26
NVIDIA unveiled new AI and simulation tools, including the Isaac Lab framework, to advance humanoid robot development at the Robot Learning Conference in Munich.
NVIDIA has announced a new suite of AI and simulation tools to accelerate humanoid robot development, according to NVIDIA. The announcement was made at the Conference on Robotic Learning (CoRL) in Munich, Germany and demonstrates NVIDIA’s commitment to empowering robotics developers.
Revolutionizing robotic learning with Isaac Lab
A highlight of NVIDIA’s announcement is the general availability of NVIDIA Isaac Lab, an open source robot learning framework. Built on the NVIDIA Omniverse platform, Isaac Lab facilitates large-scale robotics policy training applicable to a variety of robot implementations, including humanoids, quadrupeds, and collaborative robots. This framework is being adopted by leading companies and research institutions around the world, including Agility Robotics and Boston Dynamics.
Project GR00T: Promoting humanoid development
NVIDIA also introduced Project GR00T, an initiative designed to accelerate humanoid robot development. This project presents six new workflows that provide a comprehensive blueprint for overcoming challenges associated with humanoid robot capabilities, including motion generation and full-body control. Jim Fan, senior research manager at NVIDIA, highlighted Project GR00T’s collaborative efforts to advance humanoid robotics globally.
Innovative tools for developing world models
NVIDIA has unveiled the Cosmos tokenizer and NeMo Curator to help create AI representations of the world. Cosmos tokenizer improves video and image data processing through high-quality compression, while NeMo Curator simplifies video processing, significantly reducing data curation time. These tools are essential for developing accurate and efficient models of the world and are critical for robots to interact with their environment.
Commitment to the Robotic Learning Community
NVIDIA’s participation in CoRL included 23 research paper presentations and 9 workshops covering various aspects of robot learning. These contributions highlight groundbreaking advances in vision language model integration, robot navigation, and humanoid robot control. The conference served as a platform for NVIDIA to share its advancements and collaborate with the global robotics learning community.
Newly released tools and frameworks are available on platforms such as GitHub, and additional resources such as developer guides and tutorials are available for those interested in exploring NVIDIA’s innovations in robot learning and simulation.
Image source: Shutterstock