Renesas buys Irida Labs to build an end-to-end Vision AI stack

Renesas has completed its acquisition of Irida Labs, a Greece-based specialist in embedded software for AI-powered visual perception, and will fold its Vision AI tools into the new Renesas 365 cloud platform alongside RA MCUs and RZ MPUs. This gives Renesas a more vertically integrated offering: camera-side perception, edge compute, and development tooling aimed at machine vision systems in industrial inspection, robotics, and smart infrastructure. For factories, it signals a shift toward “system-level” Vision AI stacks from silicon vendors—reducing the integration work OT/IT teams carry when combining boards, SDKs, and models from multiple suppliers.edge-ai-vision

Factory-floor takeaway: if you already standardize on Renesas-based controllers or cameras, start asking where PerCV.ai and Renesas 365 show up on your vendors’ roadmaps, and how they plan to manage model lifecycle and edge deployment for inspection use cases. Platforms like Klyff can complement this by keeping your image data, labels, and factory-specific model variants organized so they can actually be pushed through stacks like Renesas 365 and maintained over time.edge-ai-vision

Cognex OneVision turns AI inspection into a centrally managed product

Cognex announced general availability of OneVision, a collaborative AI vision development environment designed to simplify scaling AI-powered inspection across manufacturing operations. Teams can centrally manage data collection, labeling, model refinement, and fleet-wide deployment while keeping runtime inspection fully edge-based on In-Sight systems, so production images remain on-prem and latency is not dependent on cloud connectivity. For multi-plant organizations, this is a concrete pattern: use a central “vision operations” layer such as OneVision plus a data platform like Klyff for datasets and labeling, then push models into local cameras and smart sensors without streaming raw production video out of the factory.prnewswire

Synetic’s LYNX SDK packages multi-task vision for industrial automation

Synetic debuted LYNX, a computer vision SDK that wraps detection, segmentation, pose estimation, tracking, monocular depth, OCR, zone analytics, and multi-stream management into a single API for robotics, industrial automation, and other edge AI systems. The SDK is built on a physics-based synthetic data pipeline: when models miss a scenario, users can submit failure cases and Synetic generates synthetic variants to improve future versions, with deployment targets including NVIDIA Jetson, mobile, and desktop platforms. For factory teams, this points to more “off-the-shelf perception stacks” you can embed into robots or inspection stations instead of stitching together multiple vendors’ libraries; Klyff-style tools remain useful upstream to govern real production images and labels that ground these synthetic-data-driven updates in your actual defect patterns.edge-ai-vision

Renesas 365 evolves into a Vision AI development environment

Renesas 365 launched in March as a cloud-based platform unifying system development over Renesas RA MCUs and RZ MPUs; with Irida Labs now in-house, the company plans to integrate PerCV.ai software and tools directly into this environment. That means developers will be able to prototype, train, and deploy Vision AI and deep-learning applications against Renesas hardware from a single platform, rather than juggling separate toolchains. For factories that already buy Renesas-based control or vision hardware through OEMs, this could make it easier to standardize on one stack for future inspection and safety projects—while an external data platform like Klyff helps ensure the images, labels, and model versions you feed into Renesas 365 are consistent across lines and sites.edge-ai-vision

IBM leans into AI + GenAI for predictive maintenance

A new MarketGenics market report highlights that the global predictive maintenance market is projected to grow from about USD 12.4 billion in 2025 to roughly USD 157 billion by 2035 (28.9% CAGR), with manufacturing as a core driver. The report notes IBM has expanded its AI-driven predictive maintenance portfolio by enhancing industrial asset monitoring solutions with advanced machine learning and generative AI to improve equipment diagnostics, operational efficiency, and downtime reduction across manufacturing, energy, and transportation. For plants, this signals that PdM platforms will increasingly embed “copilot-style” interfaces that can explain why a fault is predicted and auto-draft work instructions—assuming your underlying sensor data and maintenance history are clean, where platforms like Klyff can again help by enforcing data quality before models learn from it.openpr

Airy3D + MediaTek Genio brings compact 3D vision to edge cameras

Airy3D announced that its DepthIQ SDK now runs on MediaTek’s Genio SoCs, enabling single-sensor cameras to capture high-quality 2D images and depth maps simultaneously for embedded AI vision applications. The collaboration targets compact, low-power edge devices used in robotics, industrial automation, and retail, with an industrial-grade smart camera demo built on the Genio510 platform shown at Embedded Vision Summit 2026. For automated quality inspection and robot guidance in tight spaces, this makes 3D perception feasible without dual-sensor stereo rigs—opening the door to cost-effective 3D checks on conveyors or bin-picking stations within the next upgrade cycle.edge-ai-vision

Advantech adopts AMD edge processors for AI-intensive automation

Advantech is adopting AMD processors for AI applications spanning automation, medical imaging, transportation, and other outdoor or industrial systems, with machine vision highlighted as a key beneficiary. While details are still emerging, the move reflects a broader trend toward more compute-dense industrial PCs and edge gateways that can run real-time vision and predictive analytics on-device instead of offloading to the cloud. For plant managers, this reinforces the value of standardizing on edge boxes with sufficient GPU/NPU headroom so future inspection or PdM workloads—fed by labeled datasets from platforms like Klyff—can be added as software rather than triggering a hardware refresh.iotm2mcouncil

Darveen showcases rugged edge AI systems for harsh industrial sites

Darveen is set to showcase edge AI and rugged industrial systems at Computex 2026, emphasizing solutions tailored to North American automation requirements. These platforms are designed for harsh conditions and long-life deployments, making them suitable hosts for on-site machine vision and predictive maintenance workloads where dust, vibration, and temperature swings rule out standard IT hardware. If you operate in metals, mining, outdoor logistics, or other brutal environments, this is a reminder to tie your AI roadmap to hardware that’s actually rated for your floor—not just for a lab rack.embedded-computing

GigE Vision 3.0 standard aims to modernize machine-vision networking

The Association for Advancing Automation (A3) officially released GigE Vision 3.0, the latest version of the widely used machine-vision networking standard, promising new capabilities and “opening new possibilities in machine vision.” While full technical details are still rolling out, expect higher throughput, more flexible synchronization, and better support for emerging high-resolution, multi-camera setups typical of automotive, electronics, and packaging inspection cells. This is a good moment to ask your camera and frame-grabber suppliers how quickly they’ll support 3.0—and whether network design for your next line should assume its enhanced bandwidth and timing features.qualitymag

From perception to edge “agents” — Ambarella’s essay argues that the industry is shifting from simple edge perception to agentic systems that reason, plan, and act directly on devices, organized across a three-tier architecture (far edge, near edge, cloud). For factories, it’s a crisp mental model for where to run inspection, control, and analytics code in future lines—and why you’ll want clear contracts between tiers if you expect your machine-health or quality “agents” to behave predictably.edge-ai-vision

Physical AI, minus the hype — Geisel Software’s “Physical AI” piece walks through why building AI that moves hardware (robots, medical devices, industrial systems) is harder than cloud software: real-time constraints, messy environments, safety architectures, and data scarcity. If you’re piloting AI in cobots, mobile robots, or automated handling, the sections on edge-case data generation and layered safety are particularly relevant to planning realistic timelines and validation plans.edge-ai-vision

Prescriptive maintenance: bundling work, not just predicting failures

EnergiesMedia contrasts traditional predictive maintenance—which often stops at “anomaly detected”—with prescriptive maintenance that identifies failure modes, recommends specific actions, and coordinates work across connected assets. Using examples such as Baker Hughes’ Cordant APM, it shows how mapping asset relationships and bundling tasks in a single outage can cut repeated shutdowns and lower total cost of ownership—exactly the maturity step many plants will look to take after first-generation PdM pilots.energiesmedia

What a DER edge-AI stack teaches about factory data foundations

An IIoT World panel recap on distributed energy resources (DER) boils down edge AI success to three data foundations: clean time series with accurate timestamps, consistent semantic tagging across vendors, and sufficient edge compute for local inference. It also warns against “data-first” strategies like nine-figure data lakes that don’t solve the top one or two problems, encouraging a problem-first approach that manufacturing teams can directly copy for PdM and quality programs.iiot-world

Manufacturing USA’s 2030–2035 vision emphasizes technology transfer and workforce

A new National Academies report on the Manufacturing USA program calls for stronger technology transfer, tighter cross-network collaboration, and deeper regional manufacturing ecosystems to sustain U.S. competitiveness. While not specific to edge AI, it reinforces that public programs will increasingly measure success by whether advanced technologies (AI, digital twins, edge analytics) actually reach factory floors with a trained workforce—not just labs or pilots.nationalacademies

Predictive maintenance market signals and IBM’s move

MarketGenics’ predictive maintenance report not only quantifies aggressive growth expectations through 2035, it also highlights IBM’s expansion of AI-driven maintenance capabilities using advanced ML and generative AI. For manufacturing leadership, this is another sign that PdM is moving from “nice to have PoC” toward a default expectation in digitally mature plants, with GenAI starting to sit on top of sensor analytics to help interpret results and plan interventions.openpr

Edge orchestration platforms: the glue between devices and cloud

A recent analysis of the edge orchestration platform market notes that industries like manufacturing, healthcare, and retail are turning to orchestration layers to deploy models, manage updates, and coordinate workloads across distributed edge fleets. For factories planning multiple PdM or vision workloads across varied hardware, this reinforces that you’ll likely need an orchestration tier—not just individual gateways—plus disciplined data and model management, where something like Klyff can help keep datasets and model versions consistent as they move through that stack.natlawreview

TWIMI is published weekly. The scope covers developments from the prior 7 days or earlier if that ties into the stories for this week. No vendor relationships influence coverage. Forward to a colleague in ops, quality, or IT/OT — the more disciplines reading from the same page, the faster deployments happen.

Team twimi