Factory operations platform for AI-based monitoring and coordination

Advantech introduced what it calls an AI Factory Brain, a multi-agent system built around a factory manager agent using the NVIDIA Factory Operations Blueprint. The system is designed to monitor anomalies, identify possible root causes and coordinate responses between software agents and human operators.

The system runs on NVIDIA accelerated computing platforms and uses MIC and ICAM edge devices based on NVIDIA IGX Thor, NVIDIA Jetson Thor and NVIDIA Jetson Orin. It collects data from enterprise systems such as SAP, MES and WMS, along with edge sensors. Advantech uses NVIDIA NeMoClaw, NVIDIA Omniverse, NVIDIA Metropolis and NVIDIA Isaac Sim to support workflow automation, energy management and production analysis.

Edge AI for inspection and process monitoring

Advantech’s ICAM-540 camera and MIC-743-AT platforms use NVIDIA Jetson Orin NX and NVIDIA Jetson Thor to support AI-based automated optical inspection and local LLM or VLM chatbot deployment. These systems are intended for defect detection, production monitoring and natural language interaction in factory environments.

Physical AI for logistics and warehouse operations

Advantech’s AIR-427A platform, based on NVIDIA IGX Thor, and AFE-A702 platform, based on NVIDIA Jetson Thor, provide AI computing for forklifts, humanoid robots, autonomous mobile robots and mobile manipulation robots. These systems are designed to support navigation, robotic manipulation, material handling and coordination on the factory floor. AI agents are also intended to interpret workflow conditions, adjust production priorities and assist operators.

Security and functional safety features

Using NVIDIA NeMoClaw, including NVIDIA OpenShell runtime and policy frameworks, the architecture includes security, access control and operational governance features. The MIC-735-IT and AIR-427A platforms, based on NVIDIA IGX Thor, are also intended for functionally safe AI computing in humanoid robotics and autonomous industrial systems.

Internal pilot deployments

Advantech tested the architecture in two AI agent pilots within its own manufacturing operations. The iEnergy Agent is designed for energy management by comparing production schedules with Vision AI data and SCADA systems to control HVAC and lighting. Advantech projects a 10% reduction in total factory energy consumption after full deployment.

The Production Line Efficiency Agent uses Vision AI to collect real-time assembly line data. It analyzes productivity metrics, identifies anomalies and bottlenecks and produces recommendations and shift reports. Advantech reports that, over about six months of deployment, the system improved assembly line productivity by 12%.

Advantech plans to continue expanding its edge AI and accelerated computing portfolio for applications including smart manufacturing, automated optical inspection, industrial robotics, warehouse automation, autonomous logistics and digital twin deployments.

For more information, visit advantech.com.

The post Advantech expands NVIDIA partnership for factory AI appeared first on Engineering.com.

The IMX711 is an X-ray image sensor for inspection and measurement instrumentation which directly detects X-rays and outputs signals proportional to their energy. The new sensor offers the industry’s fastest^*1 maximum 26,100 fps high-speed imaging, achieved thanks to Sony’s proprietary circuit technology, suppressing charge saturation for accurate measurements. It also significantly reduces noise to enable improved signal detection precision in low-flux conditions, detecting differences in photon energy. It provides both high-accuracy measurements of integrated X-ray energy at a wide dynamic range and energy information acquisition at the photon level on a single sensor, a feat that has been difficult with conventional sensors. This unique feature will contribute to the advancement and diversification of X-ray inspection and measurement technologies, which are used in a wide variety of applications, from cutting-edge device inspection to scientific measurements.

*1 Among charge-integrating X-ray CMOS image sensors. According to Sony research (as of announcement on June 9, 2026).

*2 Based on the image sensor effective pixel specification method.

Examples of potential applications

• Improving precision and throughput in high-speed inspection of moving objects for battery and semiconductor applications
• Elemental mapping for distinguishing photons of different energy levels and rendering two-dimensional distribution
• Simultaneous measurement of crystal structure analysis and element analysis using photon energy information and spatial information

Main features

• High-accuracy measurements in a wide dynamic range thanks to the industry’s fastest,^*1 low-noise imaging

This product achieves the industry’s fastest^*1 maximum frame rate of 26,100 fps thanks to Sony’s proprietary circuit technology. Lowering the amount of accumulated charge per frame enables superior saturation characteristics compared to conventional sensors. At the same time, random noise, which is a technical challenge on charge-integrating sensors, has been reduced to 34 e-rms^*3 so that even faint X-ray signals are not obscured by noise and can be reliably detected. This improves measurement precision under low-flux conditions, offering photon level detection. These features enable accurate measurement of integrated X-ray energy for all pixels across low- to high-flux conditions, supporting inspection and measurement with significant differences in brightness on a single sensor, contributing to improved device throughput and an expanded dynamic range.

*3 An evaluation result. This result is calculated based on the average value of pixels in the sensor’s effective area in an environment where the sensor’s internal operating temperature is at or below 20℃. The functional guaranteed value is 60 e-rms.

• High energy resolution enabling inspection and measurement using photon energy

The new sensor uses the charge-integrating method, which makes it possible to acquire photon energy information without the need to set a threshold in advance. Furthermore, noise and signal variation are suppressed during reading to achieve high energy resolution for clear identification of differences in photon energy. Enabling acquisition of highly reliable data via high energy resolution will contribute to streamlining and improving precision in advanced inspection and measurement, which previously required several measurements in applications such as detecting differences in constituent elements at the element level and structural and material analysis to quantitatively evaluate minute state changes. It also enables post-processing under various conditions such as collecting the measurement data for all pixels, combining it with spatial information, and extracting specific energy data contributes to multifunctional inspection and measurement.

The IMX711 was developed with the collaboration between Sony Semiconductor Solutions Corporation and RIKEN. Based on a pixel structure invented by Dr. Takaki Hatsui of RIKEN, the two parties worked together on the technological development required to make it viable as a practical X-ray image sensor, including improving sensitivity and achieving high resistance to X-ray irradiation and high-voltage tolerance. Sony developed its circuit technology, manufacturing processes and packaging technology for mass production.

For more information, visit sony-semicon.com.

The post Sony readies X-ray CMOS sensor for high-speed imaging appeared first on Engineering.com.

Though summer is practically upon us and schools are soon to unleash their captive charges upon the world, research labs continue to fill journals with novel discoveries and innovations even as classrooms empty. On the subject of 3D printing in particular, here are a few recent announcements from the academic world that caught our eye.

A new platform for volumetric 3D printing

Volumetric 3D printing is one of the most exciting up-and-coming additive technologies. Similar to vat photopolymerization (VPP), also known as stereolithography (SLA), volumetric 3D printing involves using light as a catalyst to produce solid objects from liquid resin. The major difference is that while VPP/SLA are point- or layer-based, volumetric 3D printing catalyzes the entire object at once using holography or other advanced optical techniques.

The latest development in this vein comes from researchers at École Polytechnique Fédérale de Lausanne (EPFL). Building on earlier research in which they used holograms to encode 3D forms via light wave alignment to preserve laser power, the EPFL team has implemented a new platform that is reportedly 70 times more efficient compared with previous techniques.

In their experiments, the researchers used their new system to solidify entire millimeter-scale objects within seconds, and centimeter-scale objects within minutes. In addition, the method’s phase control enables holographic printing with self-healing beams, resulting in higher-fidelity 3D printed objects using light-scattering media, including those containing living cells.

“Our method’s demonstrated efficiency and precision finally makes it possible to bioprint tissue-like structures at near-clinical scale,” said Christophe Moser, head of the Laboratory of Applied Photonics Devices, in a press release. “We have printed structures substantially larger than those achieved with previous holographic approaches, despite increased light scattering caused by the embedded cells.”

In this particular study, the team used a 150-mW laser diode to print a life-sized human ear as well as a smaller scale build to confirm that embedded living cells were still viable after six days. “Our approach brings volumetric printing closer to real-scale implants, and biologically compatible manufacturing using low-power laser sources,” said lead author Maria Alvarez-Castaño in the same release.

The research is published in the journal Light: Science & Applications.

Curcumin for better 3D printed microscopic ceramics

Plenty of additive manufacturing (AM) materials companies use proprietary recipes for their products but few, if any, are likely to look in a spice rack for their next secret ingredient. Maybe they should though, based on research from Hunan University, which demonstrates that curcumin—the source of turmeric’s bright yellow color—can improve 3D printing of microscopic ceramic parts.

The natural spice works as an active, dual-purpose agent, physically screening stray printing light as well as neutralizing the free radicals that can cause print errors. According to the researchers, at a concentration of 0.01 weight percent, curcumin prevents microscopic features from blurring together without impacting build times.

The researchers report that this approach reduced the blurring error on 3D printed microscopic parts to 26.1 micrometers in laboratory tests. They also say it kept 50-micrometer holes open and free-flowing while standard resins saw them completely clogged shut.

As an added bonus, because curcumin completely burns away during the final high-temperature baking phase, the finished ceramics retain their density and mechanical strength required for their applications. According to the researchers, this presents a scalable, highly reliable pathway to the mass production of microscopic ceramic structures.

The research is published in the International Journal of Extreme Manufacturing.

3D printed speaker cover focuses audio

Engineers have been exploiting the geometric and material flexibility of 3D printing to fine-tune or dampen acoustic properties for a while now. The latest breakthrough comes from researchers at Penn State who have designed a system to manipulate sound waves so that they’re only audible in a small, targeted area.

While sound waves typically spread outward from their source, a parametric array loudspeaker (PAL) uses high-intensity ultrasonic waves to focus audible sound into a narrow, laser-like beam. While PALs have been in use for some time, they come with some significant drawbacks.

“These arrays are so directional that once the sound beam comes in contact with a surface, the sound can reflect all around the room, compromising privacy,” said lead author Jee Woo Kevin Kim in a press release. “Additionally, they struggle to produce low-end frequencies, which can take away from the experience of listening to bass-heavy music, for example.”

The solution, developed by Kim and colleagues including professor of acoustics Yun Jing, involves metamaterials. “To develop an acoustic metasurface, we use a large surface that works like a lens focusing a beam of light,” said Jing in the same release. “The surface modulates sound waves in such a way that they converge at a central point after leaving the speaker, allowing us to focus the audio into a precise area.”

The 3D printed covering can direct sound produced by the speakers into a tight “bubble” that is only audible in a space slightly larger than an inch wide and less than a quarter of an inch tall. “Using this design, our focal point is fixed in space, but the components’ passive nature allows us to substantially reduce the cost of manufacturing and implementing the acoustic metasurface,” Jing explained.

Tests showed that the PALs could effectively project frequencies as low as 38 Hz, which is close to the deepest, lowest-pitched range of human hearing, and something that traditionally requires bulky subwoofers to accomplish. Additionally, less than two inches outside the narrow target range, the researchers say, the volume drops as much as 50 decibels.

“The acoustic metasurface is about six inches in diameter — around the size of a small plate — and can be applied directly onto the surface of any PAL,” Kim said. “We believe this holds great commercial potential, as companies would just need a 3D-printer or a plastic mold to mass produce these components.”

The research is published in IEEE Transactions on Ultrasonics.

Recyclable resin could improve 3D printing sustainability

Despite many marketing claims to the contrary, AM isn’t always the most sustainable manufacturing process. However, the more ways we can find to reuse 3D printing feedstocks, the better positioned it will be to legitimately lay claim to being a green technology. Researchers from Yokohama National University have been making progress on that front with a new recyclable resin for SLA.

“Photocured 3D models cannot be recycled, so there are concerns about the negative environmental impact of discarding 3D-printed resin parts,” said co-corresponding author and professor Shoji Maruo. “Research into reusable photocurable resins is progressing, but conventional resins form irreversible cross-linked networks. As a result, previously proposed ‘recyclable’ resins either require the addition of chemical additives for reuse or rapidly degrade after one or a few reuse cycles.”

The researchers set out to investigate whether a unique property of anthracene and its derivatives could offer a solution due to its tendency to undergo photodimerization when exposed to light.

“We found that the reversible photodimerization reaction of anthracene could be a practical method for developing a truly reusable resin free from initiators, or chemicals used to induce a reaction, that can maintain performance through multiple recycling cycles while supporting high-precision stereolithography,” Maruo said.

The researchers tested their new resin using single-photon micro-stereolithography and two-photon photopolymerization and found that it behaved similarly to off-the-shelf commercial counterparts, curing via stepwise polymerization without the need for photoinitiators. “This unique feature simplifies resin formulation, eliminates contamination from additives and enables near-complete recyclability,” Maruo said. “We demonstrated that the resin could be recycled at least 10 times using two-photon lithography.”

With their proof-of-concept complete, the researchers’ next steps will be adapting the resin to a larger-scale 3D printer as well as improving its thermal response and long-term stability.

The research is published in ACS Omega.

Missed the last additive research update? Don’t worry: we’ve got you covered.

The post Additive research update: recyclable resins, musical metasurfaces, secret spices, and more appeared first on Engineering.com.

From AI assistance to autonomous engineering

The ChipStack AI Super Agent is designed to operate with a high degree of autonomy in chip design and verification workflows while still allowing engineers to inspect results, guide the process and collaborate when needed. It also integrates with collaboration environments and works with tools such as Codex and Claude Code to provide visibility into system activity and decisions.

Instead of following only step-by-step prompts, the ChipStack AI Super Agent is designed to evaluate intermediate results, determine next actions and iterate across tasks such as specification understanding, RTL generation, verification planning, formal analysis, simulation, debugging and design convergence. This allows engineers to focus more on reviewing results and setting objectives while the system handles more of the workflow.

Engineering models and production security

Cadence links autonomous agent behavior to its physics-based design and verification engines so that AI-driven actions are based on existing computational models and signoff-oriented results.

For deployment, the ChipStack AI Super Agent runs within NVIDIA OpenShell runtime, a sandboxed environment for autonomous agents that applies governance controls and is designed to protect intellectual property through policy controls, isolation and managed access to tools, infrastructure and design data. Cadence combines its design and verification engines with OpenShell security controls as part of its production deployment approach.

Expansion of Cadence agentic AI tools

Cadence said this release builds on earlier work in agentic AI with NVIDIA. After acquiring ChipStack in November 2025, Cadence introduced its first related product in February 2026 and later expanded the portfolio at CadenceLIVE in April. The company introduced ViraStack AI Super Agent for custom and analog design, InnoStack AI Super Agent for digital implementation and signoff and Cadence AgentStack as an orchestration framework for agent-based workflows across the design stack. The current update extends those capabilities further toward autonomous operation.

Availability

Cadence expects the new autonomous capabilities in the ChipStack AI Super Agent and the AgentStack orchestration framework to be available to early-access customers in the second half of 2026.

For more information, visit cadence.com.

The post Cadence expands ChipStack for autonomous design workflows appeared first on Engineering.com.

Taiwan has one of the world’s fastest-aging populations, increasing pressure on clinical staff. Managing care across clinicians, devices, treatment rooms and shift schedules remains complex, and hospital workflows often involve multiple disconnected systems.

Agentic AI is being positioned as a way to support hospital operations by coordinating clinical and operational tasks and extending workforce capacity. In this model, AI systems are used to help manage growing care demands across healthcare settings.

Specialized agents are AI systems designed for specific domains and intended to operate continuously across clinical and operational workflows in digital and physical environments. Digital agents work inside software systems to support clinical reasoning, documentation and care coordination, while physical agents such as robots and connected devices are used for logistics, monitoring and procedural support in hospital settings.

Under Taiwan’s Healthy Taiwan initiative, the region has committed $1.5 billion to build a regulated health AI ecosystem involving hospitals, academic institutions and technology companies. Foxconn is acting as an integrator among government programs, hospitals, device manufacturers and software providers to support deployment of clinical AI systems.

AI tools for clinical support

Foxconn’s CoDoctor platform is designed as a coordinated set of specialized AI agents for areas including cardiovascular care, oncology and ophthalmology. The platform is intended to help clinicians with diagnosis support, documentation and coordination across departments.

The current CoDoctor AI tools include an ECG AI agent for EKG-based screening and triage, a Corovia AI agent that reconstructs the heart and coronary arteries in 3D to shorten parts of the clinical workflow and an Endovia AI agent for colonoscopy that supports real-time lesion detection and low-latency AI-assisted workflows.

Foxconn’s clinical AI platform uses NVIDIA Nemotron open models for clinical reasoning and multimodal processing. These models use open weights, allowing healthcare organizations to manage and adapt them within their own environments.

Foxconn also introduced CoDoClaw, a clinical agent system built on NVIDIA NeMoClaw, an open-source blueprint for autonomous agent deployment, and NVIDIA OpenShell for privacy and security controls. CoDoClaw is intended to connect multiple AI tools through one clinical interface, including systems for breast cancer screening, ECG analysis, fundus imaging and coronary artery analysis.

These AI systems are being deployed at medical centers in Taiwan for uses such as ECG screening, cancer subtype classification, surgical planning and real-time support during procedures including colonoscopy.

Physical AI systems in hospitals

AI-based systems are also being used in physical hospital environments, where they operate alongside clinicians and other digital systems in wards and operating rooms.

Foxconn introduced Scrub Bot, a collaborative robot designed for surgical suites. It is intended to respond to voice commands from surgeons and adjust to changing requirements during procedures.

NVIDIA also introduced the Agent-Ready Rheo blueprint within NVIDIA Isaac for Healthcare to support development of hospital automation workflows. Foxconn is using Rheo for simulation-to-real workflows including scene reconstruction, policy training, evaluation and deployment.

Foxconn’s Nurabot, a nursing robot built on NVIDIA’s physical AI stack, completed field validation last year at Taichung Veterans General Hospital and is now moving into broader deployment. Additional rollout is under way at Taipei Veterans General Hospital, Tungs’ Taichung MetroHarbor Hospital, long-term care settings and nursing education institutions outside Taiwan.

Foxconn estimates that Nurabot can reduce the time nurses spend on transport and logistics tasks by two to three hours per day, allowing more time for direct patient care. Future Nurabot deployments are also being tested with NVIDIA NeMoClaw and physical AI technologies to support more coordinated clinical workflows and simulation-based development.

To prepare robots for hospital use, Foxconn creates digital twins of hospital facilities using NVIDIA Omniverse. These virtual models are used to test, train and validate robotic and AI systems before deployment in clinical environments. Foxconn reports that this approach has reduced deployment time by 40% and enabled navigation accuracy of 98%.

Taiwan’s healthcare AI deployment model

Taiwan currently has 85 FDA- or TFDA-cleared medical AI solutions, and major medical centers in the region are involved in AI research or clinical trials. This reflects a broad effort to incorporate AI tools into healthcare delivery.

Many medical centers in Taiwan now use AI systems in routine workflows. Deployments are active at Chang Gung Memorial Hospital, Kaohsiung Medical University Chung-Ho Memorial Hospital, MacKay Memorial Hospital, National Taiwan University Hospital, Taichung Veterans General Hospital, Taipei Veterans General Hospital and other healthcare sites that together handle more than 14 million patient encounters each year.

These projects present Healthy Taiwan as one model for countries seeking to build regulated national healthcare AI systems using NVIDIA-based infrastructure.

For more information, visit nvidia.com.

The post Foxconn deploys AI systems across Taiwan hospitals appeared first on Engineering.com.

Taiwan-based server manufacturers and other supply chain partners are building Vera Rubin-based systems for AI labs, cloud providers and hyperscale operators.

The Vera Rubin platform is a pod-scale system made up of five integrated racks for AI workloads. It combines Vera Rubin NVL72 systems, the Vera CPU, BlueField-4 networking and storage components and Spectrum-6 Ethernet racks into one platform. NVIDIA cites up to 10 times the agent throughput of the previous Grace Blackwell platform at scale.

Vera Rubin production ramp

Vera Rubin is the third generation of NVIDIA MGX rack-scale systems. The MGX design is supported by hundreds of supply chain partners, including about 150 in Taiwan, across more than 350 factories in 30 countries.

System builders, infrastructure software providers and storage companies producing Vera Rubin systems include Dell Technologies, HPE, Lenovo, Supermicro, AIC, Aivres, ASRock Rack, ASUS, Cloudian, Compal, DDN, Everpure, Foxconn, GIGABYTE, Hitachi Vantara, Hyve Solutions, IBM, Inventec, MinIO, MiTAC Computing, MSI, NetApp, Nutanix, Pegatron, Quanta Cloud Technology (QCT), VAST Data, WEKA, Wistron and Wiwynn.

Networking for large AI deployments

For scale-out and scale-across AI infrastructure, the Vera Rubin platform includes NVIDIA Spectrum-X Ethernet Photonics, a co-packaged optics switch with 200Gb/s SerDes that is now in production.

NVIDIA reports that Spectrum-X Ethernet Photonics provides up to five times the power efficiency, up to five times longer uptime and 1.3 times faster deployment than networks based on traditional transceivers.

NVIDIA positions its co-packaged optics networking for large GPU cluster deployments. CoreWeave, Lambda and Oracle Cloud Infrastructure are among the early adopters named.

The Vera Rubin platform also includes BlueField-4 DPUs with software-defined networking at speeds up to 800Gb/s and built-in multi-tenant isolation. The BlueField-4 Advanced Secure Trusted Resource Architecture is intended to simplify network operations, improve tenant isolation and increase control in large GPU clusters.

Security for AI infrastructure

AI infrastructure is increasingly being used for proprietary data, regulated content and operational models in agentic workflows. These use cases require infrastructure security for autonomous agents operating in shared or cloud environments.

The Vera Rubin platform includes full-stack confidential computing for trusted execution at rack scale. Vera Rubin NVL72 combines Vera CPUs, Rubin GPUs, NVLink networking and security features in one platform, with encryption across high-speed interconnects. Hardware-level attestation is included for system integrity.

Cloud providers identified as adopting confidential computing include CoreWeave, Firmus, GMI Cloud, IBM Cloud, IREN, Lambda, Microsoft Azure, Nebius, Nscale, SpaceXAI and Vultr.

Pod-scale security also requires software that can enforce and manage security policies across the system. The DOCA software platform is designed to provide security across Vera Rubin racks and infrastructure layers, including protection for data, agents, context memory and AI inference through BlueField-4 hardware.

DOCA supports multi-tenant isolation, zero-trust policy enforcement, runtime threat detection and end-to-end encryption at speeds up to 800Gb/s without using host CPU resources.

Infrastructure design for Vera Rubin deployments

The DSX platform provides the design and operational framework for Vera Rubin-based AI infrastructure. It combines reference designs, simulation, infrastructure software, facilities guidance and ecosystem technologies to support deployment and operation.

Built for the Vera Rubin pod architecture, DSX is intended to align system design, lifecycle management and multi-tenant operations across the stack.

Organizations adopting DSX for Vera Rubin deployments include Dell Technologies, HPE, Lenovo, Supermicro, ASUS, Foxconn, GIGABYTE, Pegatron, Quanta Cloud Technology (QCT), Wistron and Wiwynn.

For more information, visit nvidia.com.

The post NVIDIA ramps Vera Rubin for agentic AI infrastructure appeared first on Engineering.com.

The problem we’ve been living with

Let me set the stage with something you’ve probably experienced. You’re designing a collaborative robot arm for a precision assembly task. Your actuators are standard—brushless servos, maybe some harmonic drives. Good stuff. But your load path is weird. The geometry doesn’t match off-the-shelf ratios. So you either compromise on performance, redesign the mechanism, or you hit the supplier lottery and hope someone in the catalog has what you need within six months.

This isn’t unique to robotics. Across manufacturing, we’ve been working within supply chain constraints for decades. We optimize around what exists rather than optimizing the application itself. That’s not engineering—that’s constraint acceptance dressed up in a tie.

But here’s the beautiful part: that era is ending.

The hybrid supply chain

Here’s what’s changed, and it’s genuinely exciting. We now have the manufacturing technology—multi-axis CNC, additive processes, real-time production optimization—to produce custom components at near-commodity scale. Customers can specify custom gearboxes through digital interfaces, with AI-optimized tooth profiles, matched to their exact load cases and speed ratios. Manufacturing lead times are measured in weeks, not quarters.

The hybrid model works like this: you keep your standard actuators—those are your anchors, your reliability baseline. But the power transmission components? The gearboxes, the couplings, the mounting structures? Those are now customizable without cost or timeline penalties.

The economic shift is subtle but profound. A decade ago, customization meant a 3-5x cost multiplier and manufacturing lead times that made project managers weep. Now? We’re talking 1.2 to 1.5x cost, same lead time as a standard part. That changes the entire calculus of what’s possible to build.

And we’re seeing this ripple through the industry in real ways. When Tesla announced their expansion into robotic manufacturing at their Giga facilities, a huge part of their advantage came from integrating custom drivetrain components with standardized actuators—exactly this hybrid approach. They could iterate their robotic systems at the speed of software development instead of the speed of casting and machining.

AI and the optimization flywheel

Machine learning algorithms are becoming invisible middleware between design and manufacturing. You specify your load case—torque, duty cycle, space constraints, efficiency targets—and the system runs 10,000 design iterations in minutes. It explores gear tooth geometries, material selections, internal bore patterns you’d never manually consider. Then it interfaces directly with our manufacturing execution systems.

The feedback loop is genuinely exciting. We’re collecting anonymized performance data from deployed gearboxes across hundreds of customers. That data feeds back into the AI models. They get smarter. They optimize for real-world failure modes, not just theoretical textbooks. The next iteration of custom components is more robust, more efficient, and we shaved another week off manufacturing.

That’s not magic. That’s engineering and data science having a conversation.

Consider what Boston Dynamics has been doing with their humanoid robots. They’re not just hand-crafting every component—they’re using AI-driven design optimization to create actuators and power transmission systems that can handle the complex, dynamic movements these robots need. The feedback from real-world testing gets fed back into the design loop. Each generation of robot is more capable than the last, not because they’re hand-tuned by genius engineers (though those help), but because the system is learning from deployed units.

Similarly, when you look at what’s happening in collaborative robotics—companies like Universal Robots and ABB are increasingly leveraging AI to optimize custom configurations for specific applications. Customers get robots that are perfectly tuned to their workflow, not robots that are “close enough.”

IT/OT convergence: where the rubber meets the cloud

I’ll be candid: this is where most industrial companies trip. The gap between Information Technology (what runs in the cloud and talks to you) and Operational Technology (what runs on the factory floor and talks to itself) has been a chasm for thirty years.

Not anymore.

At MISUMI Americas, we’re building systems where customer design data flows seamlessly into manufacturing planning, inventory management, and logistics—all in real time. An engineer in Detroit specifies a custom gearbox, and within minutes, our digital twin has simulated the manufacturing sequence, reserved capacity, and queued the CNC programs.

What’s encouraging is that we’re seeing real momentum. Siemens’ Digital Industries division has been aggressively pushing the convergence agenda, and their success stories are multiplying. Companies are actually achieving integrated IT/OT stacks that work reliably. Standards are emerging (MQTT, OPC UA) that make interoperability viable. It’s still early, but the direction is unmistakable.

And here’s the humbling part: we’re still figuring it out. But the fact that we’re figuring it out with increasing success is what gets me excited.

Why this actually matters

Let me give you some concrete examples:

One of our robotics customers—mid-size automation integrator—was designing a palletizing system. Standard collaborative robot, but the application required a custom gear ratio that didn’t exist in catalogs. Two years ago, they would have redesigned. Added cost, longer development cycle, performance compromises.

Last year, they specified custom gearboxes through our platform. Four-week lead time. The optimization algorithm found a solution that was actually 3% more efficient than their manual design, and cost was minimal. They deployed on schedule. The customer’s line ramped production six months earlier than expected.

That’s not a huge story for the business press. But multiply that across thousands of applications, and you’re talking about the acceleration of automation adoption itself. You remove friction from the system, and adoption curves change.

Semiconductor assembly breakthrough

Another example that’s been in the news: when Nvidia ramped up their manufacturing partnerships, they needed to customize robotic assembly systems for their specific chip geometries and thermal requirements. They couldn’t wait for suppliers to develop standardized solutions. Instead, they partnered with integrators who could rapidly prototype custom gearbox solutions—designing, optimizing, and manufacturing custom components in parallel with their actuator specifications. This let them compress what would typically be 18 months of development into 6 months. That speed advantage cascaded through their production ramp.

The medical device precision play

On the medical device side, companies like Stryker and Zimmer Biomet have been increasingly deploying custom-optimized surgical robotics. Their surgical systems require precision that’s hard to achieve with generic components. By leveraging AI-driven custom gearbox design, they’ve been able to reduce backlash in their wrist mechanisms by 40% compared to previous generations, while actually reducing cost. That’s the holy grail—better performance, lower cost, through smarter design and manufacturing.

The scalability question

Here’s what keeps me up at night: can we actually scale this?

We’re building digital manufacturing infrastructure to handle thousands of custom variants without losing the economic efficiency that comes from scale. That’s the fundamental tension. The answer, I think, lies in modular standardization—keeping core components standard while allowing customization at the integration layers.

Your actuator is standard. Your gearbox is custom. Your housing adapts. Your mounting is flexible. You achieve unlimited variety from limited components.

The convergence effect

What excites me most is how these three forces reinforce each other. AI makes custom designs viable. Digital manufacturing makes them economical. IT/OT convergence makes them scalable. Each enables the others.

We’re at a moment where an engineer can:

• Specify a custom gearbox optimized for their exact application
• Have it designed by AI in hours
• See it manufactured in weeks
• Integrate it with standard actuators that have proven reliability
• Deploy it into a system that reports performance data back to improve future designs

That complete loop—from conception to learning—that’s the robotic renaissance.

What’s next

The robotic renaissance isn’t coming. It’s here. It’s happening quietly in manufacturing cells and design studios. It’s not flashy, but it’s reshaping how we think about production scalability.

What’s missing? Deeper integration standards. More engineers who understand both design and manufacturing informatics. Greater transparency in performance data. But these are surmountable challenges, and I genuinely believe we’ll solve them in the next 3-5 years.

As someone who’s been in this industry long enough to have opinions about how things used to be, I’m genuinely excited about where we’re heading. We’re moving from constraint acceptance to constraint optimization. We’re moving from “make do with what’s available” to “manufacture exactly what’s needed.” That’s engineering progress at its finest.

The future is custom components at commodity speed. And yes, that means some of my assumptions from the ‘90s are officially outdated. I can live with that. In fact, I’m celebrating it.

The post The robotics renaissance: orchestrating the future appeared first on Engineering.com.

LG and NVIDIA held a Top Management Meeting in Yeouido, Seoul, attended by Kwang Mo Koo, chairman and CEO of LG Corp., Jensen Huang, CEO of NVIDIA, and other top executives.

The meetings, attended by Kwang Mo Koo, NVIDIA CEO Jensen Huang, and heads of major companies, expand the scope of mid- to long-term strategic cooperation leading industrial innovation in the AI era.

The expanded collaboration combines NVIDIA’s cutting-edge AI technologies with LG’s manufacturing and infrastructure capabilities to implement AI most quickly and comprehensively in customers’ daily lives and industrial sites. The collaboration between LG, which possesses decades of manufacturing innovation know-how and vast life data assets accumulated through customer touchpoints around the world, and NVIDIA, is expected to accelerate global AI innovation across both industry and daily life.

1. Cooperation in physical AI and robotics

The two companies will pursue a win-win strategy in the field of Physical AI, spanning manufacturing to robotics, by maximizing synergies based on each company’s core capabilities.

LG has accumulated production technology data and know-how from global manufacturing sites, while NVIDIA provides NVIDIA Isaac, Omniverse and Cosmos AI and simulation technologies.

By combining these strengths, the two companies plan to enhance AI-driven manufacturing competitiveness, build an autonomous manufacturing ecosystem in which the entire process from raw material procurement to production, logistics, and customer delivery is connected in real time through data and AI, and establish it as a new global smart factory standard.

The two companies will also pursue development of NVIDIA’s next-generation robot foundation model GR00T, further strengthening cooperation in robotics.

LG and NVIDIA will advance robot development capabilities and performance through strategic cooperation across the entire robotics field, from data collection and generation, simulation, training, and actions, in a wide range of robots including humanoids and logistics robots.

LG Innotek will serve as the eyes and ears of robots based on its world-class optical technologies, and will develop high-performance sensing modules and optical components optimized for NVIDIA AI infrastructure.

Furthermore, LG CNS is building an ecosystem that enables anyone to easily adopt AI robots in manufacturing and logistics sites. By integrating NVIDIA’s robotics technologies, including NVIDIA Isaac open robotics frameworks, Cosmos open world models and Isaac GR00T open robotic foundation models, into its industrial robot platform, “PhysicalWorks,” the company is accelerating the AI transformation of logistics and manufacturing floors.

2. Cooperation in AI factories

The two companies will also expand cooperation in the field of next-generation AI Infra (AIDC), which will support the AI era.

LG Electronics will further enhance its AI Infra capabilities by collaborating with NVIDIA on cooling solutions for AI Infra thermal management, including coolant distribution units (CDUs) and cold plates, as well as on prefabricated modular design technologies aligned with NVIDIA’s DSX reference design.

LG Energy Solution will develop 800V direct current (DC)-based data center power solutions with NVIDIA to deliver power efficiently for next-generation AI factories. Through this, the company aims to improve the energy efficiency of AI data centers and take the lead in the next-generation data center energy market.

LG CNS plans to build next-generation AI data centers with improved scalability and energy efficiency by adopting the NVIDIA DSX AI factory reference design, while LG Uplus plans to build a large-scale AI Infra utilizing NVIDIA Rubin GPUs.

3. Cooperation in mobility

Together with NVIDIA, LG will accelerate the realization of safer and more intelligent autonomous driving and SDVs.

LG Electronics will advance mobility AI systems, including next-generation advanced driver assistance systems (ADAS), by integrating its proprietary in-vehicle infotainment (IVI) capabilities with NVIDIA’s autonomous driving platform NVIDIA DRIVE Hyperion.

LG Innotek will expand development of core automotive components optimized for the NVIDIA DRIVE Hyperion architecture, including communication modules, sensing solutions, and automotive lighting systems.

4. Strengthening technology alliance to expand the EXAONE ecosystem

LG and NVIDIA will also strengthen their technology alliance to enhance Korea’s AI competitiveness.

LG AI Research plans to improve training efficiency and inference performance in the development of EXAONE by utilizing NVIDIA Blackwell GPUs, along with its AI development platform, NVIDIA Nemotron, NVIDIA NeMo, and inference performance enhancement software, NVIDIA TensorRT-LLM.

LG also plans to expand its adoption of NVIDIA-powered AI agents across the LG Group, including LG’s enterprise AI agent service ChatEXAONE, thereby cooperating to accelerate enterprise AI transformation.

For more information, visit lgcorp.com.

The post LG and NVIDIA widen ties across AI, factories and mobility appeared first on Engineering.com.

With the rise of AI technology, GIGABYTE integrates AI-driven tuning, hardware design, and advanced BIOS optimization for a smarter, faster, and more adaptive motherboard platform. At the core of GIGABYTE’s AI-enhanced innovation is X3D Turbo Mode 2.0, designed to maximize AMD Ryzen X3D processor performance through a dynamic AI overclocking engine and a dedicated onboard hardware chip. The exclusive OC engine is trained on massive datasets and optimized for per-processor tuning, while the onboard chip continuously monitors platform conditions and workload behavior in real time. This intelligent hardware-software integration enables adaptive optimization for AI computing, multitasking, and gaming scenarios.

GIGABYTE also redefines DDR5 memory performance with D5 Bionic Corsa technology. The AI Snatch Engine analyzes trained tuning datasets in real time to optimize DDR5 memory and CPU performance with a single click.  The advanced PCB engineering employs advanced AI algorithms throughout the motherboard to enhance signal integrity and maintain peak performance across multi-layer motherboard layouts. In addition, HyperTune BIOS, GIGABYTE’s AI-driven BIOS optimization technology, intelligently fine-tunes memory behavior to achieve significant memory clock speed and enhanced overall system performance.

By combining AI-enhanced tuning technologies with advanced motherboard engineering, GIGABYTE continues to shape the future of gaming and AI computing platforms. Other than performance-focused models, GIGABYTE also offers a diverse range of cable stealth and elegant wood edition models to suit gamers, creators, and AI enthusiasts.

For more information, visit gigabyte.com.

The post GIGABYTE adds AI tuning features to motherboard lineup appeared first on Engineering.com.

Wrong-way driving remains one of the most dangerous roadway safety challenges, often resulting in fatal collisions. ALTOS-WAY has been developed to provide transportation agencies with a reliable, easy-to-deploy detection solution that identifies dangerous vehicle movements quickly enough to enable timely warnings and response.

Built on the Opsys ALTOS platform, ALTOS-WAY combines LiDAR sensing, embedded computing, and embedded perception software in a single roadside device. The system continuously tracks vehicles in three dimensions, directly measuring position, velocity, and direction of travel to identify wrong-way movements and generate immediate alarm events for integration with roadside warning systems, traffic management platforms, and other transportation infrastructure.

ALTOS-WAY performs detection and alarm generation directly on the sensor. The system operates 24/7, uses standard roadside mounting and Ethernet infrastructure, and requires only a single Power-over-Ethernet connection for power and communications.

Built on Opsys’ true solid-state LiDAR architecture, ALTOS-WAY contains no moving parts, helping to reduce maintenance requirements and lifecycle costs while supporting long-term roadside operation. The solid-state design eliminates mechanical failures associated with sensing systems, enabling higher reliability in demanding outdoor environments. The underlying ALTOS platform is designed for a 160,000-hour MTBF, includes a standard three-year warranty with optional five-year coverage, and is compliant with NEMA TS2, and Build America, Buy America (BABA) requirements.

ALTOS-WAY is available immediately through Opsys and its distribution partner, Gades Sales Co., Inc.

For more information, visit opsys-tech.com.

The post Opsys launches real-time LiDAR wrong-way detection system appeared first on Engineering.com.