NVIDIA Jetson Orin Nano 2 brings physical AI to drones and robots 英伟达Jetson Orin Nano 2将物理AI带入无人机和机器人

TL;DR

• NVIDIA has launched the Jetson Orin Nano 2, an entry-level edge computing board designed to run generative AI and physical AI workloads directly on devices like drones, robots, and vision systems without relying on cloud infrastructure.
• The board delivers 78 TOPS of AI compute, 8GB of memory, and an eight-core Arm CPU, offering twice the inference performance of the Jetson Orin Nano Super while using 40% less power at 15 watts.
• NVIDIA's core argument is that small and medium AI models have now reached accuracy levels previously only achievable by large frontier models, making edge deployment of language and vision-language models viable.
• Early partners include Wing (Alphabet's drone delivery subsidiary), Matic Robots, Cognex, and Doosan Bobcat, with over three million developers already building on NVIDIA's Jetson ecosystem.
• The board supports memory-efficient inference of models such as NVIDIA's own Cosmos and Nemotron, as well as Gemma 4 and Qwen 3, running on NVIDIA's open software stack and Jetson agent skills.

Why It Matters

This launch signals a pivotal shift in AI deployment strategy: frontier-level model performance is no longer exclusive to data centers, and edge hardware is now capable of running real-time generative AI workloads. For AI practitioners and robotics developers, this means the barrier to deploying intelligent, perception-driven systems in resource-constrained environments has significantly lowered. It also reinforces NVIDIA's dominant position in the physical AI and edge computing market, setting a new baseline for what entry-level edge hardware can achieve.

Technical Details

• Compute and Architecture: The Jetson Orin Nano 2 features 78 trillion operations per second (TOPS) of AI compute, 8GB of memory, and an eight-core Arm CPU, all within the same compact form factor as its predecessor.
• Performance Gains: It delivers 2x the inference performance of the Jetson Orin Nano Super, attributed to improved Tensor Cores and higher memory bandwidth. At 15 watts, it consumes 40% less power while matching the Super's performance level.
• Model Support: The board is optimized for memory-efficient inference of large language models (LLMs) and vision-language models (VLMs), including NVIDIA's Cosmos and Nemotron, Google's Gemma 4, and Alibaba's Qwen 3.
• Software Ecosystem: Runs on NVIDIA's open Jetson software stack with Jetson agent skills, enabling developers to deploy real-time reasoning, perception, and navigation capabilities on edge devices.
• Target Applications: Designed for drones (delivery and inspection), home robots, vision AI systems, carrier boards, and reference designs, with partner hardware from AAEON, ADLINK, Advantech, and others.

Industry Insight

• The convergence of model efficiency gains and edge hardware advancement is accelerating the "physical AI" wave, where robots and autonomous systems can reason and perceive in real time without cloud dependency. Developers should evaluate edge-deployable model variants (quantized, distilled) to leverage this hardware effectively.
• NVIDIA's strategy of anchoring its edge platform to a vast developer ecosystem (3M+ developers) creates a strong moat. AI professionals building robotics or drone applications should consider the Jetson ecosystem for its mature tooling, partner hardware options, and proven deployments like Wing's drone fleet.
• The emphasis on power efficiency (40% reduction at equivalent performance) makes this board particularly relevant for battery-constrained platforms like delivery drones and consumer robots, where thermal and energy budgets are critical design constraints.

TL;DR

• NVIDIA发布Jetson Orin Nano 2边缘计算板，定位入门级物理AI开发平台，面向无人机、机器人和视觉系统
• 提供78 TOPS AI算力、8GB内存和8核Arm CPU，推理性能达前代Jetson Orin Nano Super的两倍
• 15瓦模式下功耗降低40%，支持在边缘端实时运行大语言模型和视觉语言模型
• 合作伙伴包括Wing（Alphabet无人机配送）、Matic Robots、Cognex等，推动家庭机器人和无人机应用落地

为什么值得看

这篇文章标志着边缘AI硬件的重要里程碑——前沿模型性能已能部署到入门级边缘设备，为物理AI应用提供了可行的硬件基础。对AI从业者和硬件开发者而言，这展示了端侧推理的可行性路径，以及边缘计算在降低延迟和功耗方面的战略价值。

技术解析

Jetson Orin Nano 2搭载78 TOPS AI算力、8GB内存和8核Arm CPU，采用改进的Tensor Cores和更高内存带宽，推理性能达到前代产品的两倍。在15瓦模式下，功耗较Orin Nano Super降低40%，同时保持同等性能水平。

硬件支持NVIDIA开放软件栈和Jetson agent skills，可运行Cosmos、Nemotron、Gemma 4和Qwen 3等语言模型和视觉语言模型，专为内存高效边缘推理优化。

首批合作伙伴包括Wing（Alphabet无人机配送子公司）、Matic Robots（家用清洁机器人）、Cognex和Doosan Bobcat，覆盖无人机配送、家庭机器人、视觉AI系统等多个应用场景。

行业启示

边缘AI正从云端向终端迁移，小模型性能已接近一年前的前沿模型水平，这为实时推理和物理AI应用打开了新空间。硬件厂商需要关注低功耗、高性能的边缘计算方案，以支持无人机、机器人等对功耗敏感的应用场景。

开发者生态的构建至关重要——NVIDIA已拥有超过300万开发者，通过开放软件栈和参考设计降低开发门槛，加速物理AI应用的商业化落地。