Back to blog

Advantech Puts CODESYS EtherCAT Motion on Jetson Edge AI

Advantech has combined NVIDIA Jetson edge AI, CODESYS SoftPLC and EtherCAT motion control on MIC-AI platforms. We examine where the architecture helps—and the real-time, safety and lifecycle checks...

Advantech has announced an industrial edge-computing architecture that combines NVIDIA Jetson processing, CODESYS SoftPLC functions and EtherCAT motion control on the same MIC-AI platform. The November 19, 2026 announcement is aimed at physical-AI machines that need vision or inference close to the process while still coordinating deterministic field devices.

What Advantech has integrated

The architecture is offered on two different computing profiles. The MIC-742-AT uses NVIDIA Jetson Thor and provides one QSFP28 interface, one 5GbE port, four USB 3.2 ports and industrial I/O. The MIC-732-AO uses Jetson AGX Orin, includes dual CAN interfaces and can be configured for GMSL camera acquisition. Advantech says both can run a real-time Linux environment with CODESYS SoftPLC and an EtherCAT master, allowing AI inference, control logic and coordinated motion to reside on one industrial computer.

In the demonstration described by Advantech, a MIC-742-AT ran NVIDIA Isaac for Manipulation, Isaac ROS 2 and a real-time motion stack while driving a multi-axis robot over EtherCAT without an external conventional PLC. CODESYS supplied the IEC 61131-3 control environment and EtherCAT master functions, including distributed-clock synchronization and device diagnostics.

Why the combined platform matters

Many robot and machine-vision cells currently split the workload among an AI computer, a PLC or motion controller, and one or more industrial network gateways. Consolidation can reduce data copies and simplify the path between a perception result and a motion command. It can also make synchronized image acquisition, path planning and drive feedback easier to observe from a single system.

That does not mean every machine should remove its PLC. The value is strongest where the AI workload is part of the machine cycle and latency between perception and action matters. A packaging line that only sends quality images to an analytics server may gain little from convergence. A robot that changes its trajectory after detecting an object, by contrast, may benefit from keeping inference and motion orchestration close together.

What the demonstration does not establish

Advantech's announcement proves that the software stack can be assembled and operated on the MIC hardware, but it does not publish a deterministic cycle-time benchmark, maximum axis count, jitter figure or validated drive list for a production machine. It also does not turn the Jetson platform into a safety controller. Emergency-stop, safe torque off and other safety functions still need a documented safety architecture using appropriately certified components.

Engineers should therefore treat the launch as an integration option, not as a blanket replacement for established motion platforms. For comparison, installed EtherCAT machines may already use dedicated controllers such as the Omron NX701-1620 motion controller or the Omron NJ501-1500. These links illustrate the type of installed equipment a modernization project may encounter; they do not imply compatibility with the Advantech/CODESYS stack.

Engineering checks before adoption

  • Real-time partitioning: confirm how CPU, memory and network resources are reserved so peak GPU inference load cannot disturb the EtherCAT task.
  • Network design: keep the EtherCAT segment distinct from high-bandwidth camera and enterprise traffic. If the machine uses branches, document every junction and propagation delay; devices such as the Omron GX-JC06 EtherCAT junction show why topology is a hardware decision, not merely a software setting.
  • Thermal margin: validate enclosure airflow, ambient temperature, dust loading and GPU derating under the actual inference workload.
  • Recovery behavior: test cold start, corrupted model files, camera loss, EtherCAT link interruption and watchdog behavior. Define which functions remain available if the AI application stops.
  • Lifecycle ownership: assign responsibility for JetPack, Linux, CODESYS, robot libraries and trained-model updates. A patch process that is acceptable for analytics may be too disruptive for a production controller.
  • Cybersecurity: separate engineering access from runtime services, sign or verify deployable artifacts, restrict container privileges and record changes to both control code and AI models.

Procurement and maintenance implications

The commercial question is not only the price of one industrial computer versus several controllers. Buyers should compare engineering licenses, supported software versions, long-term image availability, spare-unit restoration time and the skills required to maintain a converged Linux/PLC/AI stack. A spare device is useful only if the plant can reproduce its operating system, CODESYS application, EtherCAT configuration, AI model and accelerator dependencies.

Advantech has previously targeted edge-AI integration in other form factors; our earlier analysis of the ASR-D501 platform highlights the same need to separate headline compute capability from environmental and integration checks. For machine builders, the new CODESYS/EtherCAT option is technically significant because it moves that discussion directly into motion control. The final decision should follow measured real-time tests on the intended drives, cameras and safety architecture.

Source: Advantech, “Advantech Integrates NVIDIA Jetson-Powered Computing with CODESYS EtherCAT Motion Control for Physical AI,” published November 19, 2026. Official announcement. Editorial analysis by PLC ProTech, November 20, 2026.

Leave a comment

Please note, comments need to be approved before they are published.