NVIDIA Jetson is a family of system-on-module (SoM) and developer kit edge computing platforms that combine NVIDIA GPU cores with ARM-based CPU clusters, purpose-built for deploying deep learning inference, computer vision, and robotics workloads at the edge with constrained power budgets. Modules in the Jetson family — including Nano, TX2, Xavier, and Orin — span from entry-level embedded devices to high-performance autonomous machine platforms, all running NVIDIA’s JetPack SDK which provides CUDA, cuDNN, TensorRT, and ROS integration.
Content
- NVIDIA launched the first Jetson TK1 module in 2014, targeting robotics researchers who needed GPU-accelerated perception outside the data centre. The platform grew through successive generations: Jetson TX1 and TX2 brought higher-efficiency Maxwell and Pascal architectures; Jetson Xavier (AGX and NX) introduced NVIDIA’s Volta GPU with dedicated deep learning and vision accelerators (DLA and PVA); and Jetson Orin (2022) delivered Ampere GPU cores with up to 275 TOPS of AI performance in a 15-60 W power envelope.
- Every Jetson module ships with JetPack, a comprehensive SDK layering Ubuntu Linux, CUDA, cuDNN, TensorRT, DeepStream, and Isaac ROS on top of the SoC. TensorRT graph optimisation and INT8 quantisation allow large neural networks to run at real-time frame rates without cloud offload. The modules expose camera CSI interfaces, PCIe, USB 3, and GPIO, enabling integration with depth cameras, lidars, IMUs, and actuators in robotics platforms. NVIDIA’s Jetson containers on NGC provide pre-built Docker images for popular frameworks including PyTorch, TensorFlow, and Triton Inference Server.
- The significance of Jetson lies in collapsing the edge-cloud AI gap: autonomous mobile robots (AMRs), surgical assistance systems, smart retail cameras, and drone-based inspection all use Jetson hardware to perform multi-modal sensor fusion and policy execution locally, avoiding latency and connectivity constraints of cloud inference. The platform has become the reference hardware for many open-source robotics stacks, and Isaac ROS builds NVIDIA-optimised perception nodes atop ROS 2.
- In 2024-2025, Jetson Orin dominates the high-performance edge tier while Jetson Nano 2 and Orin Nano address cost-sensitive IoT deployments. NVIDIA’s broader robotics initiative — including Project GR00T for humanoid foundation models — explicitly targets Jetson as the deployment substrate. Competitive pressure from Qualcomm’s Snapdragon platforms and purpose-built NPU ASICs is growing, but NVIDIA’s software ecosystem and CUDA lock-in maintain strong market position in robotics and smart city applications.