NVIDIA has recently introduced the NVIDIA Jetson AGX Thor developer kit and production modules, innovative robotics computers intended for a wide range of industries such as manufacturing, logistics, transportation, healthcare, agriculture, and retail. Key industry players like Agility Robotics, Amazon Robotics, Boston Dynamics, Caterpillar, Figure, Hexagon, Medtronic, and Meta have already embraced this technology, while others like 1X, John Deere, OpenAI, and Physical Intelligence are evaluating its potential to enhance their physical AI capabilities. The CEO of NVIDIA, Jensen Huang, highlighted that the Jetson Thor was developed for the multitude of developers engaged in creating robotic systems that interact with and influence the physical world. He emphasized its unparalleled performance, energy efficiency, and ability to support various generative AI models at the edge, making it a crucial component in the era of physical AI and general robotics.
The NVIDIA Jetson AGX Thor is described as the ultimate platform for next-generation robotics, equipped with an NVIDIA Blackwell GPU and 128GB of memory, capable of providing up to 2,070 FP4 teraflops of AI compute power while operating within a 130-watt power limit. Compared to its predecessor, the NVIDIA Jetson Orin, the Jetson Thor offers up to 7.5 times higher AI computing performance and 3.5 times greater energy efficiency, enabling the seamless execution of diverse generative AI models. This system-on-module addresses a significant challenge in robotics by facilitating the operation of multi-AI workflows, enabling robots to engage intelligently with people and their environment in real-time scenarios. It allows for real-time inference, crucial for high-performing physical AI applications, including humanoid robotics, agriculture, and surgical assistance.
Notably, the Jetson Thor is supported by the comprehensive NVIDIA Jetson software platform, catering to physical AI and humanoid robotics, capable of accommodating various AI frameworks and generative models. It seamlessly integrates with NVIDIA’s software suite from cloud to edge, including tools like NVIDIA Isaac for robotics simulation and development, Isaac GR00T for humanoid robot foundation models, NVIDIA Metropolis for vision AI, and NVIDIA Holoscan for real-time sensor processing. The Jetson platform and NVIDIA’s robotics ecosystem have attracted over 2 million developers and a network of 150+ hardware system, software, and sensor partners since its establishment in 2014. Jetson Thor is extending this frontier, particularly for visual AI agents and intricate robotic systems such as humanoids and surgical robots, drawing the interest of renowned technology leaders in the robotics domain.
Leading industry voices recognize the significance of NVIDIA’s Jetson Thor in advancing their robotics capabilities. For instance, Brett Adcock, CEO of Figure, highlighted the critical role of running powerful AI models directly on robots for real-time learning and interaction, which the Jetson Thor facilitates with its server-class performance in a compact and energy-efficient design. Similarly, Tye Brady, Chief Technologist at Amazon Robotics, emphasized the necessity of computational power and energy efficiency offered by the Jetson Thor for developing AI-enabled robots that can operate autonomously and intelligently in dynamic environments.
NVIDIA has made the Jetson AGX Thor developer kit available for purchase, starting at $3,499, while the Jetson T5000 production modules can be obtained from global distribution partners. Production systems and carrier boards are also accessible through embedded partners. NVIDIA, a pioneer in accelerated computing, remains committed to advancing technology and driving innovation in the robotics and AI space with its cutting-edge solutions like the Jetson Thor.
About the Author
