High-Powered Vision Controller At The Edge Of Production

Cognex has introduced the In-Sight 6900 vision controller, an AI-ready industrial computing platform designed to bring advanced machine vision processing directly to the factory floor. Powered by NVIDIA Jetson technology, the compact controller supports transformer-based AI architectures, multi-camera synchronization, and inspections at resolutions up to 65 MP. The system integrates with Cognex cameras, optics, and lighting while supporting industrial protocols such as EtherNet/IP, PROFINET, and OPC UA. Combined with Cognex’s OneVision cloud-based software and AI inspection tools, the In-Sight 6900 is built to streamline deployment, reduce training time, and enable more complex inspection and defect-detection applications at the edge.

The In-Sight 6900 Vision controller. Image used courtesy of Cognex

The In-Sight 6900 Vision Controller

A typical vision system from Cognex consists of a camera with a lens and a lighting system. These combined components are capable of a wide range of vision inspections. By adding a vision controller, image inspection is offloaded to a faster, more powerful computing platform. Using a vision controller allows for high-speed inspections, lighting and camera synchronization, reduced operational costs, and an optimized vision process. The In-Sight 6900 is designed with AI in mind, offering the ability to run transformer-based architectures, 256 GB of onboard storage, and NVIDIA Jetson software, which is compatible with all Jetson AI modules. Each controller supports up to 65 MP resolution with a widened field of view to cover large and small inspections. Four high-bandwidth Ethernet ports are provided to connect up to four cameras that can be synchronized together. The 6900 also ships factory-ready, supporting popular industrial communication protocols such as Ethernet/IP, ProfiNet, and OPC UA. For lighting and accessories, the 6900 features 4 digital inputs, 8 outputs, and 2 USB-C ports for expandable storage.

The 6900 with some example accessories. Image used courtesy of Cognex

AI Vision Tools

Vision inspections often require teaching many different images iterations to build a library of what defines a good or bad part. Using AI, we can significantly reduce this teaching time. The In-Sight controller is designed to operate advanced AI-based vision tools, such as the Edge AI tool, which uses pre-trained models for defect detection, OCR, classification, and assembly verification. Advanced AI tools enable deep customization of inspection tools using larger datasets and multiple learning models to support more complex inspections. All these tools can be configured and taught using Cognex’s OneVision or In-Sight vision suite.

A front view of the 6900 showing all the ports. Image used courtesy of Cognex

Vision Software

Engineers and designers use advanced software to train AI vision models; this software often runs on the controllers, as it’s expensive to license and requires powerful hardware. Cognex’s OneVision is a cloud-based software that can be accessed remotely and used to teach AI models and keep vision projects up to date. Applications can be built in one place and deployed to multiple installations, saving time and money with re-deployments. Pre-defined workflows provide fast, efficient ways to develop vision applications while still allowing customization.

As industrial computers become more of an industry standard, vision companies like Cognex are taking advantage and building controllers capable of advanced inspections without the need for large, bulky servers. As these controllers become more affordable, will this mean the death of the all-in-one vision system?