NVIDIA showcased the Jetson platform as an embedded environment for building AI and robotics applications in the real world, with modules and development kits that can be transported and used in classrooms, laboratories and innovation spaces. The presentation appeared in a video featuring Sarah Guo, founder of venture capital firm Conviction and co-host of the No Priors AI podcast, explaining how the platform combines computing power with portability.
NVIDIA says Jetson modules and development kits are designed to run robots, autonomous machines and edge AI projects, with support for advanced open models. The Jetson Device Skills and Jetson BSP Skills tools also allow students, researchers and developers to use software agents to help write and optimize code and deploy applications at the edge.
Jetson Orin Nano Super for Early Projects
NVIDIA presents the Jetson Orin Nano Super as a practical path for developers new to robotics, combining capabilities for generative AI, computer vision, agent building and prototyping in a small development kit. Its AI performance reaches 67 trillion operations per second (TOPS).
The post showcases several examples of its use, including the SidewalkPilot model, which autonomously performs maneuvers in a modified electric ride-on car for children, and the voice-and-vision Reachy Mini Jetson Assistant, which runs locally on the Reachy Mini Lite robot without the cloud, API keys or an internet connection during operation. Another example demonstrates building an AI-powered robot using an open-weight Mistral model.
Jetson AGX Orin for More Complex Work
The Jetson AGX Orin delivers up to 275 trillion AI operations per second and targets advanced robotics projects, autonomous machines and applied research. According to NVIDIA, it can be used for computer vision, generative AI, autonomous navigation and industrial automation.
Examples cited by the company include a web interface for streaming real-time inference from a vision-language model through a camera feed, as well as the SMoRes system developed by a robotics team at Carnegie Mellon University to create a 3D map of the environment and search for survivors in time-sensitive rescue scenarios.
Jetson AGX Thor for Larger Workloads
The Jetson AGX Thor targets the next generation of humanoid robots and autonomous systems that require advanced real-time inference at the edge. The module provides up to 2,070 teraflops of AI performance under the FP4 benchmark, along with 128GB of memory.
NVIDIA showcases the Matcha Bot experiment, carried out by the SIGRobotics team at the University of Illinois Urbana-Champaign, using two robotic arms running the NVIDIA Isaac GR00T N1.5 model to autonomously prepare matcha. The company also presents the Multimodal AI Studio experiment, a development environment that combines vision, audio and language inputs for conducting multimodal AI experiments at the edge.
The company points to a livestream series comprising three educational modules for running generative AI, building arm-control agents, and using vision-language and vision-language-action models in physical AI applications on Jetson.