GlobalSpec recently had the opportunity to ask Szymon Chawarski, product line manager at Teledyne Vision Solutions, about the growing role of AI in machine vision systems and cameras. Chawarski is the perfect person to ask, with 15 years of experience in computer vision, automation and product design. He has recently focused on applying AI and deep learning to solve challenging real-time inspection applications and on growing adoption of AI in automation.

GlobalSpec (GS): At Automate 2025, Teledyne is showcasing the BOA3. How might edge-AI change the way engineers design machine vision systems in production and manufacturing?

Szymon Chawarski, product line manager, Teledyne Vision SolutionsSzymon Chawarski (SC): BOA3 runs AI models on-camera, allowing engineers to deploy AI on the production line without installing and maintaining PCs with GPUs in an industrial environment. This keeps costs down and reduces system complexity. One of the main goals of edge-AI is to empower novice users and allow them to train vision systems without much programming experience. This can enable engineers to design vision systems that are more generalized, allowing the end users to fine-tune the inspection to their specific part or task. We’ve paired our BOA3 camera with our Astrocyte AI trainer software to allow end users to easily train and deploy specialized AI models.

GS: What types of industries and applications will BOA3 lend itself to?

SC: BOA3 and its onboard software is designed to be very flexible and solve many machine vision use cases. It can be customized and tailored to almost any application. We sell into any industry that does automated manufacturing. Our biggest customers are in automotive, electronics and packaging, but we sell to almost any industry including many recognizable name brand manufacturers.

GS: How can engineers leverage BOA3’s edge-based AI capabilities to minimize latency and bandwidth demands compared to traditional cloud or server-based processing methods?

SC: With an edge device such as BOA3, engineers are trading pure AI horsepower of server-based systems for the simplicity and ease-of-support of local inference. In manufacturing, many applications only require simple AI models, which can be tuned to run real-time on local processors. After they are set up, edge AI devices can run effectively without any internet connections, without any subscription or service fees, and with very low power consumption. The trade-off is that the models and applications solved are simpler, but there is definitely a large market for this type of solution.

GS: As more AI-enabled vision systems are deployed, how should engineers approach lifecycle management and system integration?

SC: In most ways, AI-enabled vision systems have the same lifecycle and integration loop as traditional systems. Proper lens and light selection still help with accuracy. IO, communications, data-logging and user interface design are all important factors. Both types of systems need to pass validation and factory acceptance tests, and the methods used to test both should be similar.

The unique and critical aspect of AI systems is gathering, storing and categorizing sample images. AI models always need high-quality image data to train on. Engineers need to spend a bit more effort gathering samples and saving images for AI training datasets, but the tradeoff is that once the image datasets have been created, the AI training normally takes on all the heavy lifting for the vision programming.

GS: What other edge AI solutions from Teledyne will be at Automate 2025? Or what’s next on the development roadmap?

SC: We will also be showcasing our PC-based Sherlock machine vision software. Sherlock is an industry-leading no-code machine vision software development environment. It can run AI models on an edge PC and is optimized to run inference on GPU or CPU. Sherlock is a fully capable machine vision software and it can combine AI, traditional vision tools and also 3D profiler measurements to create comprehensive vision systems. It has a powerful set of algorithms packaged in an approachable programming environment.

GS: What are you most looking forward to at Automate 2025?

SC: Interest in machine vision has been exploding in recent years, mainly driven by advances in AI. Automate is my favorite show because there is such a wide range of technologies being shown all in one place, with many exhibitors showing off their machine vision and AI solutions. I’m excited to see what new features and products are being brought to the market this year.

GS: What are your best show tips or advice?

SC: Take time to walk the show! Every year there are many interesting and fun demos showing cool new robotics and automation ideas.