Aetina Corporation has begun shipping the DeviceEdge AIE-KT78 and AIE-KT68, a pair of edge AI computers built around NVIDIA's Jetson Thor platform and aimed at collaborative robots, humanoid robots, and other autonomous machines. Both systems pack NVIDIA Blackwell-architecture AI compute, high-bandwidth sensor connectivity, and real-time industrial motion control into a single enclosure just 80mm thick, so a robot's perception, decision-making, and actuator control can run on one board instead of several.
The AIE-KT78 runs NVIDIA's Jetson T5000 module with 128GB of 256-bit LPDDR5X memory and delivers up to 2,070 FP4 TFLOPS. The AIE-KT68 steps down to the Jetson T4000 module, 64GB of memory, and up to 1,200 FP4 TFLOPS, a lower-power, lower-cost option for integrators who don't need the top tier of throughput. Both run the full NVIDIA stack, including JetPack 7, CUDA, TensorRT, DeepStream, Holoscan, and Isaac ROS, and can run vision-language-action models directly on the device. For engineers evaluating robotic arms or autonomous mobile platforms for the shop floor, on-device VLA inference means a machine can read a scene and generate a motion plan without a round trip to a remote server, cutting the latency that has kept some perception-to-action pipelines out of production.
On the I/O side, the AIE-KT Series adds QSFP28 connectivity, delivering up to 4x25Gbps depending on configuration, alongside dual RJ45 10GbE ports and support for up to eight GMSL2 camera inputs, enough bandwidth to pull in simultaneous feeds from high-resolution cameras, 3D LiDAR, radar, depth cameras, and IMUs. A separate RJ45 1GbE port runs as an independent EtherCAT Master with microsecond-level synchronization, connecting the AI inference pipeline directly to motors, joints, and actuators. Aetina says the setup removes the need for a second controller dedicated to motion, a common integration bottleneck in cobot and humanoid robot programs.
Both systems are rated for a 9-48VDC input range and operation from -25°C to 55°C, with USB 3.2, isolated digital I/O, and M.2 expansion slots. "The competitive edge for cobots and humanoid robots has shifted from the compute performance of a single model to whether a system can integrate perception, reasoning, decision-making, and action in real time in real-world environments," said Richard Hung, vice president of Aetina's Product Division. Hung said the company, an NVIDIA Elite Partner, backs the systems with mechanical design, thermal engineering, and system integration support meant to move developers from proof-of-concept to production faster.
The AIE-KT78 and AIE-KT68 are available to ship now, according to Aetina.
