NVIDIA Jetson AGX Thor Launch: Preparing Jetson Thor Compatible Cameras for Next-Generation Embedded Vision

The Ultimate Platform for Humanoid Robotics

Introducing NVIDIA Jetson Thor, the ultimate platform advancing robotics with generative reasoning, multimodal processing, and integration into next-generation robots. NVIDIA today announced the Jetson AGX Thor™ Developer Kit, a new class of robotic computing platform delivering unmatched performance and scalability for humanoid robotics, autonomous machines, and other physical AI applications. As the latest high-performance platform for advanced physical AI and humanoid robotics, NVIDIA Jetson Thor showcases NVIDIA’s leadership and innovation in the field. Jetson AGX Thor belongs to a new class of robotic computers designed specifically for next-generation humanoid robots, supporting advanced generative AI models, sensor integration, and real-time processing. As part of NVIDIA’s family of robotic computers, Jetson AGX Thor is built for autonomous machines and next-generation humanoid robots.

Built on the NVIDIA Blackwell GPU architecture with 128 GB of memory, Jetson AGX Thor delivers up to 2070 FP4 TFLOPS of AI compute within a 130 W power envelope. Compared with its predecessor, Jetson AGX Orin™, Thor provides up to 7.5x higher AI compute and 3.5x greater energy efficiency.

Key features of the Jetson AGX Thor Developer Kit include the NVIDIA Blackwell GPU architecture, 128 GB of memory, up to 2070 FP4 TFLOPS of AI compute, and a 130 W power envelope.

The Developer Kit offers a complete prototyping platform, while the new Jetson T5000 production module delivers the same performance in a scalable form factor for integration into end products. Together, they provide developers a clear path from concept to production.

Jetson AGX Thor at a Glance

Jetson AGX Thor combines massive compute with a rich set of accelerators and connectivity designed for real-time, multi-sensor applications:

• 2070 FP4 TFLOPS AI compute with the NVIDIA Blackwell GPU and transformer engine, delivering industry-leading AI performance for AI and robotics applications
• 128 GB LPDDR5X memory with 273 GB/s bandwidth
• 14-core Arm® Neoverse®-V3AE CPU with MIG (Multi-Instance GPU) technology for parallel and scalable AI workloads, including Blackwell Multi Instance GPU support
• 3rd-generation Programmable Vision Accelerator (PVA), dual encoders/decoders, and optical flow accelerator
• Connectivity for sensor fusion: QSFP (4x25GbE), wired Multi-GbE RJ45, PCIe Gen5, USB 3.2, CAN, HDMI/DP, and more, with multiple USB ports and carrier board/carrier boards for flexible hardware integration
• Power configurable between 40 W and 130 W
• Part of the NVIDIA Jetson platforms family, supporting a wide range of edge AI and robotics applications, and compatible with existing humanoid robot platforms
• Includes the NVIDIA Jetson T5000 module as a production-ready option

The system software includes the NVIDIA AI software stack and a comprehensive AI software stack for robotics and generative AI, supporting rapid development and deployment. Jetson AGX Thor also supports NVIDIA Metropolis for visual agent AI, NVIDIA Holoscan for high-performance AI sensor processing, and the Holoscan Sensor Bridge for high-speed sensor fusion and high-speed sensor integration.

This architecture enables developers to run multiple generative AI models in real time – from Vision Language Action (VLA) models like NVIDIA Isaac™ GR00T N for humanoids, to popular large language models (LLMs) and vision-language models. Jetson Thor runs the latest generative AI models and provides AI compute to effortlessly run and deploy them at the edge, enabling real-time AI, AI inference, and video analytics for intelligent visual agent workers. The platform supports building AI agents and deploying them at the edge, with advanced video search and video analytics capabilities for visual agent workers across industries.

From Prototyping to Deployment

The Developer Kit is designed for evaluation and prototyping, with extensive I/O and tethering options that make it easier to integrate into humanoid robot platforms or automation testbeds. For production, the Jetson T5000 module delivers the same compute, memory, and I/O capabilities in a compact, scalable system-on-module (SoM).

This approach allows developers to:

• Validate multi-sensor applications on the Developer Kit.
• Transition seamlessly to the T5000 module for deployment in robotics, industrial, medical, or smart space environments.

Bringing FSM:GO to Jetson Thor

With Jetson AGX Thor now available, FRAMOS is preparing FSM:GO camera modules to work seamlessly with the new platform. For developers, this will mean:

• Prototype faster with pre-focused, lens-matched camera modules ready to connect to Jetson.
• Simplify integration by combining industrial-grade imaging with Thor’s compute power for real-time applications.
• Open-source drivers available on GitHub, making it easier to get started and customize solutions without barriers.

By extending FSM:GO compatibility to Jetson Thor, FRAMOS is focused on making it easier for developers to move quickly from concept to production while building reliable, high-performance vision systems.

Looking Ahead

The launch of Jetson AGX Thor marks a significant milestone in edge AI computing. Its combination of Blackwell-powered compute, advanced accelerators, and robust software stack is designed to accelerate the development of humanoid robotics and other demanding physical AI applications.

At FRAMOS, we are committed to supporting this platform with imaging solutions that scale from 2D color FSM:GO modules to 3D depth sensing with the D400e family, alongside ISP tuning and Holoscan-ready integration for maximum image quality.

As Jetson AGX Thor and the T5000 module become available, FRAMOS will provide ongoing updates on FSM:GO compatibility and resources to help developers move from concept to production with confidence.