Industrial vision projects keep stalling on the same mechanical problem: four camera streams, nowhere to put a GPU box, and a vehicle or AMR electrical environment that kills consumer SBCs. Aetina Corporation’s answer, carried in SecurityInformed’s industrial automation newswire, is a pair of palm-sized DeviceEdge systems—AIE-VN34/44 and AIE-VO24/34—built around NVIDIA Jetson Orin modules and four native GMSL2 camera ports.

Aetina positions the DeviceEdge AIE-VN/VO family for smart transport, AMR, and line-side inspection where cabinet volume is already gone.

The Orin NX variants (8 GB / 16 GB) are rated up to 100 TOPS; Orin Nano Super Mode units (4 GB / 8 GB) reach about 67 TOPS. Both live in a fanless 136.3 × 132 × 63 mm shell—small enough to hide beside a battery pack or inside an operator console that was never designed for a 2U inference server.

Four GMSL2 cameras without a spaghetti harness

Each box exposes four GMSL2 Fakra-Z ports for long-reach, low-latency camera links. Aetina’s validated adapter boards and camera modules are billed as plug-and-play out to roughly 15 meters of image transmission, which is the difference between a tidy vehicle roof loom and a custom LVDS science project. Surrounded-view monitoring, blind-spot detection, AMR obstacle avoidance, and in-line defect inspection are the named jobs—exactly the multi-camera workloads that overwhelm USB camera trees on a generic PC.

Four Fakra-Z GMSL2 ports are the product’s practical differentiator versus USB-camera edge PCs.

James Ho, manager of Aetina’s Edge Computing Product Division, framed the launch around production reality: multi-camera perception is becoming the sensing layer for AI agents on vehicles, AMRs, and industrial equipment, and customers want hardware that is field-rugged and mass-production ready on day one—not a lab carrier board with a promise of a future enclosure.

Palm-sized Jetson Orin systems are meant to sit next to the process, not in a remote IT closet.

Rugged numbers that matter on a forklift and a line

Both platforms passed MIL-STD-810H shock and vibration testing and carry E-Mark (E24) automotive electronics certification. Operating range is listed from −25°C to +55°C with 9–36 V DC input and ignition power control for clean vehicle start/stop. Isolated CAN FD plus GPIO/RS-232 tie into vehicle and machine I/O; GbE, USB 3.2, HDMI, M.2 storage, and optional M.2 wireless cover the usual integration checklist. Software ships on NVIDIA JetPack 6.2 BSP today, with JetPack 7.2 support planned.

Aetina’s NVIDIA Elite Partner BSP story is part of the mass-production pitch, not only the TOPS headline.

On a factory floor, that package reads less like a security appliance and more like a compact vision co-processor you mount beside a cell that already runs a PLC or PAC. The PLC still owns interlocks and motion; the DeviceEdge owns multi-camera inference and returns results over Ethernet or CAN. Plants standardizing spare vision and control hardware often keep peer industrial brands—such as Honeywell sensing and control SKUs—in the same storeroom so the edge box is not an orphan platform.

Units are listed as available for mass production now. The evaluation work that remains is application-specific: lens and GMSL2 camera selection, thermal soak in the real enclosure, and a clear contract for how inference alarms land in the existing PLC tag list before the first AMR leaves the pilot aisle.