NVIDIA Jetson Thor helps general robotics and physical AI to reason in real-time
Robots have the chance to become more intelligent as physical AI developers use NVIDIA Jetson Thor modules, new robotics computers that can serve as the brains for robotic systems in research and industry
Robots require rich sensor data and low-latency AI processing, and real-time robotic applications need AI compute and memory to deal with concurrent data streams from various sensors.
The improved performance that JetsonThor can offer will enable roboticists to process high-speed sensor data and perform visual reasoning at the edge. This development can lead to new possibilities for multimodal AI applications like humanoid robotics. Jetson Thor can also be used for applications such as surgical assistants, smart tractors, delivery robots, industrial manipulators and visual AI agents, with real-time inference on the device for more complex AI models.
Advancements for real-time reasoning
Jetson Thor is designed for generative reasoning models, enabling the next generation of physical AI agents to run in real-time at the edge while reducing cloud dependency.
Jetson Thor supports all popular generative AI frameworks and AI reasoning models, thanks to its optimisation with the Jetson software stack, which enables the low latency and high performance required for use in the world.
AI frameworks and reasoning models include Cosmos Reason, DeepSeek, Llama, Gemini, and Qwen models, as well as domain-specific models for robotics, such as Isaac GR00T N1.5. This allows any developer to experiment and run inference locally.
The NVIDIA CUDA ecosystem supports Jetson Thor throughout its lifecycle to enhance its throughput and generate faster responses with future software releases.
Jetson Thor modules also run the full NVIDIA AI software stack to accelerate nearly every physical AI workflow, with platforms including NVIDIA Isaac for robotics, NVIDIA Metropolis for video analytics, AI agents, and NVIDIA Holoscan for sensor processing.
These software tools enable developers to build and deploy applications, such as visual AI agents that analyse live camera streams to monitor worker safety, humanoid robots capable of performing manipulation tasks in unstructured environments, and smart operating rooms that can guide surgeons based on data from multi-camera streams.
Advancing Research Innovation
At Carnegie Mellon University's Robotics Institute, a research team is using NVIDIA Jetson to power autonomous robots that can navigate complex, unstructured environments to perform tasks such as medical triage and search and rescue.
Sebastian Scherer, an associate research professor at Carnegie Mellon and head of AirLab, said: “We can only do as much as the compute available allows. Years ago, there was a big disconnect between computer vision and robotics because computer vision workloads were too slow for real-time decision-making — but now, models and computing have gotten fast enough so robots can handle much more nuanced tasks.”
