NVIDIA has recently unveiled its new-generation robotics computer, the Jetson Orin Nano 2. Compared with its predecessor, the Jetson Orin Nano Super, the new model delivers 2× higher generative-AI inference performance, along with a 40% reduction in power draw for equivalent workloads under the 15 W operating mode. Boasting 78 TOPS of sparse INT8 AI compute, the module features outstanding energy efficiency for use cases including robots, delivery and inspection drones, and vision AI systems.
On the hardware front, the Jetson Orin Nano 2 retains the 260-pin SO-DIMM form factor measuring 69.6 × 45 mm, with pin-to-pin compatibility with existing Orin Nano carrier boards to lower hardware migration costs. It integrates an 8-core Arm Cortex-A78AE CPU, a refined Ampere GPU and next-generation Tensor Cores, paired with 8 GB LPDDR5x memory to deliver up to 78 TOPS sparse INT8 throughput, doubling inference performance versus the Orin Nano Super. Power consumption is dynamically adjustable across the 15- 40 W range. At 15 W, it achieves equal performance with 40% less power than the prior generation, making it ideal for battery-powered mobile intelligent devices. There is no on-board eMMC storage; boot options include NVMe M.2 and microSD. Full multi-channel 4K video encode-decode engines are retained for concurrent data processing from multiple vision sensors.
For software, the NVIDIA Jetson Orin Nano 2 offers full compatibility with the NVIDIA Software stack. Powered by CUDA, cuDNN and TensorRT-LLM, it supports local deployment of large-language and multimodal vision models such as NVIDIA Nemotron, NVIDIA Cosmos, Gemma4 and Qwen3. Deep integration with the Isaac ROS robotics framework enables end-to-end edge workflows covering perception, environment understanding, LLM inference and motion decision-making. Generative-AI capabilities are brought onto robot endpoints to reduce heavy reliance on cloud connectivity. This is highly valuable for robots and unmanned inspection units, enabling real-time semantic comprehension and autonomous decision-making even offline.
As an NVIDIA Elite Partner, TWOWIN Technology has kicked off pre-research on robotics solutions built around the Jetson Orin Nano 2. Work will cover robot system-level adaptation targeting humanoid robots, wheeled inspection robots and low-speed mobile robots. Tasks include carrier-board bring-up, system image tuning, multimodal model porting and sensor driver adaptation. Leveraging improved multi-core CPU performance and enhanced LLM-inference energy efficiency, TWOWIN Technology will develop lightweight robotics reference designs to help developers accelerate prototype validation and shorten product development cycles.
Driven by the rapid growth of the physical-AI industry, the Jetson Orin Nano 2 fills the gap for on-device generative-AI compute in entry-level robots by balancing compute power, power draw and cost.TWOWIN Technology will proceed with sample testing and make evaluation access plus technical support available for industry developers to accelerate real-world deployment of next-generation edge robots.
