Edge AI solution provider Aetina has launched its DeviceEdge AIE-KT78 and AIE-KT68 flagship edge AI systems.
The new systems are designed to support local multimodal generative AI, large language model (LLM), vision-language model (VLM) and vision-language-action (VLA) model execution, while connecting high-resolution sensors with motors, joints and actuators through high-bandwidth interfaces and EtherCAT control.
Powered by NVIDIA Jetson Thor, both systems integrate NVIDIA Blackwell architecture’s high-performance AI compute, high-bandwidth sensor connectivity, and deterministic industrial control into a compact edge system.
The system is purpose-built for collaborative robots (Cobots), humanoid robots and next-generation autonomous machines. Aetina said the system can help developers perceive, reason, decide and act in unison in real-world environments.
As robots evolve from fixed-routine automation toward context-aware, autonomous decision making, there is a rapidly escalating compute demand. Existing edge platforms that deliver only a few hundred TOPS are therefore struggling to keep up.
“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 Hun, vice president of product division at Aetina.
AIE-KT78 is powered by the NVIDIA Jetson T5000 module with 128GB of 256-bit LPDDR5X memory, delivering up to 2,070 FP4 TFLOPS of AI performance. The AIE-KT68, meanwhile, is powered by the NVIDIA Jetson T4000 module with 64GB of 256-bit LPDDR5X memory, delivering up to 1,200 FP4 TFLOPS.
Both systems support on-device execution of multimodal generative AI, LLM, VLM and VLA models, reducing cloud latency and strengthening data control.
The systems enables robots to run advanced VLA models locally, integrating visual perception, language understanding, and action generation into a single decision-making pipeline that interprets environmental context and responds in real time.
To handle such data-intensive perception workloads, the AIE-KT Series simultaneously integrates QSFP28 connectivity (delivering up to 4x25Gbps of bandwidth depending on configuration) and dual RJ45 10GbE ports. This enables high-speed processing of massive real-time data streams from high-resolution cameras, 3D LiDAR, radar, depth cameras, IMUs, and various industrial sensors, while supporting up to 8 GMSL2 camera inputs to provide the high-bandwidth data foundation robots need for multi-view environmental perception and spatial understanding.
A dedicated RJ45 1GbE EtherCAT port functions as an independent EtherCAT Master, using microsecond-level synchronisation to tightly connect high-level AI inference with underlying motors, joints, sensors, and actuators.
For collaborative robots, this architecture enhances the flexibility to safely work alongside people and quickly switch between tasks. For humanoid robots, it supports more complex spatial awareness and continuous motion decision-making.
“The AIE-KT78/68 transforms the powerful compute capability of NVIDIA Jetson Thor into a compact, mass-production-ready system,” Hung said.
“As an NVIDIA Elite Partner, we combine over a decade of edge AI engineering experience with a fully validated peripheral ecosystem, providing end-to-end technical support – from mechanical design and thermal engineering to BDP and system integration – to help developers and system integrators reduce integration risk and total cost of ownership, accelerate the path from proof-of-concept to scaled deployment, and further realise the return on their automation investment.”
The architecture also reduces the need for additional control hardware and cross-platform integration, simplifying the system architecture for cobots, humanoid robots, industrial robotic arms, and advanced autonomous machines.
The AIE-KT78/68 integrates compute performance, industrial interfaces, and thermal design into a compact system just 80mm thick, supporting a wide 9–48VDC input range and stable operation across −25°C to +55°C. The system is equipped with USB 3.2, isolated digital I/O, and M.2 expansion slots.
It runs Linux with the full NVIDIA AI software stack, including NVIDIA JetPack 7, NVIDIA CUDA, NVIDIA TensorRT, NVIDIA DeepStream, NVIDIA Holoscan, and NVIDIA Isaac ROS, helping developers optimise and deploy multimodal inference, computer vision, sensor fusion, and robotics applications.
