What's Shipping
Basler's glass defect detection
Specialized for micron-level defects in glass substrates and Through Glass Vias (TGV) in a single scan. This is critical for quality control in advanced electronics manufacturing where component miniaturization demands extreme precision.
As semiconductors and micro-components become smaller and more complex, these ultra-high-precision inspection capabilities become essential for ensuring reliability. Link
MVTec HALCON 25.05
Three significant updates: Deep 3D Matching for robust object detection using 2D images, Deep OCR with alignment-based text correction, and QR code rectification that works on curved surfaces.
This demonstrates the maturation of deep learning applications to effectively handle real-world, imperfect data - exactly the kind of messy conditions that exist on actual factory floors. Link
Airy3D DepthIQ™ on Qualcomm Dragonwing
The DepthIQ software development kit now supports Qualcomm's Dragonwing platforms. This makes advanced 3D vision capabilities more accessible for embedded robotics and industrial IoT devices, standardizing depth sensing on power-efficient hardware that's already deployed in many autonomous mobile robots and factory systems.
When sophisticated capabilities become available on commodity platforms, adoption accelerates rapidly - we've seen this pattern before with machine learning inference moving from specialized chips to standard processors. Link
The AI Giants Are Paying Attention
Microsoft's process manufacturing report
80% of manufacturers surveyed are using or planning generative AI. More interesting than the adoption rate: they're not just optimizing existing processes but using AI to design new materials and predict molecular behaviors. That's upstream innovation, not just factory floor efficiency. Link
Meta's open source push
Released two vision models: one for detailed video analysis (Meta Perception Language Model) that can understand complex actions and how they're performed, and another for 3D object localization. The video model's ability to analyze the nuances of how tasks are executed could have applications in process optimization and quality assessment. Link
The 3D localization model (Locate 3D) is particularly relevant - accurate 3D object positioning is a foundational requirement for robotic grasping, precision assembly, and navigation in unstructured or semi-structured manufacturing environments. Making these capabilities open source could accelerate the development of more flexible industrial robots. Link
Google's Project Astra advances
At I/O, Google demonstrated significant improvements to Project Astra, their multimodal AI assistant that can understand and interact with the world through cameras. The most compelling demo showed real-time problem-solving guidance - helping someone fix a bike by visually analyzing the issue and walking through repair steps. For manufacturing, this points toward AI assistants that can provide real-time troubleshooting and process guidance directly on the factory floor, without workers needing to consult manuals or call experts. Link
The Startup Angle
Lumafield's Triton CT system
Production-ready CT scanning that delivers complete internal inspections in 5 seconds per part with browser-based analysis requiring no specialized training. This moves non-destructive testing from expensive lab environments to real-time production floors. Link
Emmi AI
Austrian startup trying to speed up physics simulations with AI. Traditional simulations are too slow for real-time optimization. If they can crack this, it means faster design cycles and more "what-if" scenarios for process optimization. Link
CuspAI
Using generative AI to design new materials. This is upstream from manufacturing but could be huge - imagine AI discovering lighter, stronger, or more sustainable materials that change what's possible to build. Link
What It Actually Means
Three trends are converging: AI is making machine vision more accessible, hardware is getting more flexible, and open source is democratizing advanced capabilities. The result is that sophisticated vision inspection is moving from "enterprise only" to "anyone with a real need."
But there's a catch. The Microsoft report identifies data quality as the "biggest obstacle to generative AI adoption." All these shiny new tools don't matter if your data foundation is a mess. Smart manufacturers are fixing their data infrastructure first, then adding the AI layer.
The open source movement (Meta's models, various SDKs) means smaller players can access capabilities that used to require massive R&D budgets. That's going to shake up the competitive landscape - in a good way.
The Reality Check
For all the AI excitement, companies like OPT are still launching "laboratory-grade accuracy" lenses. Because no amount of AI can fix bad input data. The fundamentals still matter - you need good optics and clean data before the smart algorithms can help you.
Need Help with Machine Vision?
If you're evaluating machine vision for your facility, we'd love to chat. Elementary builds vision systems that actually work in production - no PhD required. Reach out at sales@elementaryml.com.
.svg)
