NVIDIA and Alphabet’s Intrinsic are joining forces to advance autonomous robotic manipulation, unveiling their collaboration at the Automate trade show in Chicago. The partnership involves integrating NVIDIA AI and Isaac platform technologies into Intrinsic’s systems to significantly improve robotic grasping capabilities and industrial scalability. The project features NVIDIA’s Isaac Manipulator, which was introduced at GTC in March. Isaac Manipulator includes foundation models and modular GPU-accelerated libraries, enabling industrial automation companies to build efficient workflows for complex manipulation tasks through accelerated AI training and task reprogramming.
Foundation models, based on transformer deep learning architecture, allow neural networks to understand relationships within data. They are trained on extensive datasets and are adept at processing and interpreting robot information. This results in advanced robotic perception and decision-making, enabling zero-shot learning, where tasks can be performed without prior examples.
The collaboration highlights a promising step towards a versatile robotic grasping skill applicable across different grippers, environments, and objects. Wendy Tan White, CEO of Intrinsic, emphasized in a blog post that this partnership exemplifies how foundation models can simplify large-scale processing challenges, reduce development costs, and increase flexibility for end users. She is set to elaborate further on the implications of AI for innovation and growth during a keynote at Automate.
NVIDIA’s Isaac Sim, built on the Omniverse platform, allows Intrinsic to generate synthetic data for vacuum grasping using CAD models of sheet metal and suction grippers. This data is then used to develop a prototype for their customer, Trumpf Machine Tools, which specializes in industrial machine tools. The prototype employs Intrinsic’s Flowstate platform, offering an AI-based developer environment for visualizing processes, perception, and motion planning. Using Isaac Manipulator’s grasp pose generation and CUDA-accelerated robot motions, this simulation provides cost-effective evaluation before real-world deployment.
NVIDIA and Intrinsic aim to expand modular AI capabilities for robotic arms, offering a comprehensive collection of foundation models and GPU-accelerated libraries to boost productivity across a wider range of robotics and automation tasks.
