Advantech MIC-742 — Jetson Thor Edge AI System, 2,070 FP4 TFLOPS

Specifications

Overview

The Advantech MIC-742 is an industrial edge AI system built on NVIDIA Jetson Thor, delivering up to 2,070 FP4 TFLOPS of AI compute in a deployable robotics computer. Advantech positions it specifically for humanoid and general-purpose robotics, where the workload is dominated by transformer-based perception and vision-language-action (VLA) models rather than conventional convolutional pipelines. The MIC-742 can be paired with the NVIDIA Holoscan Sensor Bridge to create an ultra-low-latency sensor-to-inference path, keeping raw camera and sensor data on a real-time pipeline instead of round-tripping through system memory and software stacks.

Key Benefits

Datacenter-class AI at the robot: 2,070 FP4 TFLOPS lets an edge robot run the same transformer and VLA models that previously required an off-board GPU server, removing network latency and connectivity dependence from safety-relevant control loops. Deterministic sensor pipelines: the Holoscan Sensor Bridge and GMSL-class camera aggregation deliver an ultra-low-latency path from sensor to inference, which is the limiting factor for closed-loop manipulation and navigation. Software already written: the platform runs NVIDIA Isaac ROS, Isaac GR00T and the Advantech Robotic Suite, so perception, VSLAM, pose tracking and manipulation stacks can be brought up without inventing a middleware layer. Industrial packaging: Advantech builds the MIC series for factory, warehouse and mobile robotics environments rather than lab benches, with industrial power input and mounting.

Applications

Humanoid and collaborative robots, autonomous mobile robots (AMR/AGV), machine vision and inspection cells, warehouse and logistics automation, agricultural and construction autonomy, and multi-camera visual AI agents that need natural-language query over live video.

Related Products