Jetson AGX Thor Developer Kit

An advanced edge AI computing platform powered by the cutting-edge Blackwell GPU architecture, designed specifically to bring real-time control, generative AI and multi-modal transformer models directly to mobile robotics. Engineered to be the "brain" of the next generation of autonomous machines, it packs immense computational density into a compact form factor suitable for embedded hardware projects.

Poised towards physical and collaborative workflows, where other human operators might interact with untethered vehicles (warehouse and logistic robots), humanoid robots and robot arms - and any complex machine that could benefit from full-stack autonomy on the edge. Applications don't end there, use these machines for developing these systems, assisting in manual workflows like manufacturing, product inspection, and safety.

Historically, relying on cloud computing to handle complex robotic perception and reasoning introduced crippling latency - often upwards of 100 milliseconds, making dynamic, real-time control near humans unsafe and unreliable. The Thor solves this by running Vision-Language-Action (VLA) models entirely locally. By processing all sensor data on the edge, it can achieve system latencies of under 10 milliseconds, allowing untethered humanoid robots and complex manipulators to instantly perceive, reason, and act without ever waiting for a server response.

Features

• Massive Generational Leap: Compared to the NVIDIA Jetson AGX Orin, it provides up to 7.5x higher AI compute and 3.5x better energy efficiency.
• Comprehensive Suite of Accelerators: Includes a third-generation Programmable Vision Accelerator (PVA), dual encoders and decoders, an optical flow accelerator, and more to handle specialised processing tasks.
• Execute large multi-modal foundation models on the edge: Thanks to 128GB of high-speed onboard memory, eliminating the memory bottlenecks that previously prevented advanced autoregressive processing on mobile vehicles.
• High-Speed Sensor and Network Capability: The SoM supports up to 4× 25GbE connections, and a rich I/O set including HDMI, DP, USB 3.2/USB-C, and CAN for high-bandwidth and low-latency real-time sensor ingestion, processing, and fusion.
• Extreme Tensor Performance & Real-Time Responsiveness: Powered by a Jetson T5000 module with the Blackwell GPU (2560 cores and 96 Tensor Cores) and a 14-core Arm® Neoverse®-V3AE CPU. It delivers up to 2070 FP4 / 1035 FP8 TFLOPS - allowing your robots to run complex generative models (LLM, VLA like NVIDIA® Issac™ GR00T N, ViT) locally without any cloud dependency.
• Deploy deep intelligence on battery-operated platforms: By leveraging the module's highly efficient 130W maximum power envelope, maximise runtime for untethered autonomous systems.
• From Prototype to Production: Each NVIDIA development kit is built off an easy-to-integrate System on Module. If you need more camera ports or Ethernet, there are design files available to spin your own custom carrier board.

Specifications

* Refer to the Software Features section of the latest NVIDIA Jetson Linux Developer Guide for a list of supported features.
** Low-speed I/O specification is subject to change.

Resources

• Datasheet
• Getting Started
• NVIDIA Blog Post
• Documentation - In-Depth Specs, Carrier Board Design Guide

Product Includes

• NVIDIA Jetson T5000 module with heat sink and reference carrier board (Pre-installed)
• DC power Supply 140W
• WiFi Module (Pre-installed)
• 1 TB NVMe (Pre-installed)
• Quick Start Guide