IN Brief:
- An AIE-VX28/18-IP demonstration processes eight camera streams with NVIDIA Jetson AGX Orin and DeepStream for loading-area monitoring.
- The rugged AIE-VX28/18-IP family supports eight GMSL2 inputs, four M12 PoE cameras, 10GbE and IP66 protection.
- JetPack 7.2 enables a specified 241 TOPS mode on Jetson AGX Orin 32GB, compared with 200 TOPS in the earlier mode.
Aetina will demonstrate industrial edge AI systems using NVIDIA Jetson hardware at Smart City Expo World Congress in Barcelona. Its programme includes an eight-camera loading-area safety application, factory inspection, compact vehicle computers and adaptive street lighting. The demonstrations connect rugged processing hardware to the particular sensors, power supplies and installation conditions found in factories and mobile equipment.
In the loading-area example, camera feeds from a truck reach an AIE-VX28/18-IP computer running NVIDIA DeepStream on CUDA. The application identifies the worker nearest the vehicle while handling eight concurrent image streams locally. The same perception approach is proposed for forklifts, excavators and other off-highway machines. Its performance in an installation depends on the camera view, lighting, occlusion, model configuration and the speed at which decisions must be returned.
The in-vehicle hardware offers eight GMSL2 camera inputs, four M12 Power over Ethernet camera connections and 10GbE networking. GMSL2 transports image data from connected cameras across a serial interface, while the powered Ethernet ports can serve cameras using a different cabling and supply arrangement. The interfaces allow designers to match sensors and harnesses to the vehicle layout; eight live streams in the demonstration do not imply that every possible connected camera operates at the same resolution and frame rate.
Aetina specifies IP66 protection and a fanless enclosure for its AIE-VX28/18-IP family, with a 9–48 V DC supply range, operating temperatures from −25°C to 70°C and E-Mark certification. These specifications address dust and water ingress, vehicle supply variations and outdoor operation. The family includes separate Jetson AGX Orin 64GB and 32GB configurations, so compute capability and available memory depend on the installed module.
Software configuration influences how much of that hardware can be used for inference. Aetina’s planned JetPack 7.2 demonstrations include Ubuntu 24.04, Linux kernel 6.8 and CUDA 13, together with tools for examining memory reserved by bootloaders, the kernel and unused services. Any reclaimed region has to be assessed against the allocations needed for system operation; a nominal increase in free memory does not automatically increase model execution speed.
For the Jetson AGX Orin 32GB configuration, Aetina identifies a Super Mode rating of 241 TOPS compared with 200 TOPS in the earlier mode. Those figures represent the stated peak operations capability for a specific platform configuration. Real inference rates depend on numerical precision, network architecture, memory transfers and the processing work required by the cameras. The eight-stream loading-area demonstration describes a practical workload alongside that processor specification.
At a produce sorting demonstration, the camera pipeline must deliver classification results in time for downstream handling. Aetina will pair a MegaEdge AIP-FR68S with an NVIDIA GPU to grade tangerines during sorting and processing. The visual classifier has to receive images of the passing produce and associate each result with a particular item before a sorter can act on it. Conveyor speed, triggering and the arrangement of the inspection station determine how the inference result fits into production.
Smaller vehicles and autonomous mobile robots can instead use Aetina’s AIE-VN34/44 and AIE-VO24/34 systems. Based on Jetson Orin NX or Orin Nano modules, these compact devices offer up to four GMSL2 inputs and selected configurations rated at up to 100 TOPS. These previously introduced models will be shown in Barcelona in mobile equipment applications. Their smaller size changes the available space for cameras, connectors and thermal management.
Roadside cameras also feed Aetina’s adaptive street lighting demonstration. A CoreEdge MXM module fitted with NVIDIA RTX PRO 500 Blackwell Embedded graphics can process road imagery to inform lighting adjustments, while radar or lidar can supply inputs where optical cameras are unsuitable. Aetina will present these systems at the 3–5 November event, showing how sensor placement, environmental protection, available processing and electrical power differ across each deployment.
