12 June 2026 | Interaction | By Editor Robotics Business NEWS <editor@rbnpress.com>
As manufacturers accelerate their adoption of AI-driven automation, the industrial edge is becoming a critical hub for intelligence, decision-making and autonomy. In this exclusive interview with Robotics Business News, Gene Juknevicius, Senior Solution Architect at Emerson, and Manish Sharma, Global Marketing and Business Development Leader at Emerson, discuss the growing role of Physical AI, edge computing, autonomous operations and cybersecurity in shaping the next generation of industrial automation.
What strategic gap in industrial automation were you aiming to solve through the partnership with SiMa.ai?
Industrial PCs (IPCs) are targeted at the industrial edge. That means they need to be designed to operate in harsh environments—vibration, shock, extreme temperatures, dirty conditions—while still bringing powerful compute capability. In most cases, those compute capabilities can handle AI workloads; however, AI is evolving rapidly, and some of the most cutting-edge technologies drive technology platforms to their limits. In cases where the IPC hardware itself does not have the compute capability necessary, such as in a vision tool with many simultaneous video streams and a need for extremely low latency and high frame rate, the system might need external hardware to accelerate those capabilities.
The traditional choice would be to bring in a graphics processing unit (GPU). But a GPU card dramatically reduces the ruggedization of the IPC. With a GPU, there’s a fan inside, temperature ranges are reduced, and the lifecycle of the equipment is shortened. The SiMa.ai Machine Learning System on Chip (MLSoC) technology is an industrial-quality product. Adding the card to our IPC lets us increase the processing power without reducing the ruggedization specifications. Users can tackle the most powerful AI applications right at the edge, which is quickly becoming a critical differentiator for competitive advantage as industries truly embrace Industry 4.0 and start adding AI to processes and, someday, closed-loop control.
How do you see Physical AI transforming traditional factory operations over the next five years?
In the short term, we’re already seeing improvement in areas like quality assurance. Today quality assurance is implemented by taking a specific case, applying a specific model, and relying on that model to identify deviations before the end of production, thereby reducing rejects and helping manufacturing and quality assurance teams adjust manufacturing parameters in real time. Going forward, as we can add more local large language model (LLM) capabilities, we can reduce much of the pretraining, deploying models smart enough to take on more capabilities across multiple areas. The need for training a specific model for a specific problem is likely to diminish, making way for more universal use cases.
As soon as we can run these models locally, taking advantage of LLMs on the edge devices without needing to go to the cloud, it starts to become a better option for various applications. In effectively improving what we can do at the edge, we reduce cloud computing and dependency, while eliminating latency, security issues, and additional cost of sending data offsite for processing. That will create new opportunities for quality control, predictive analytics, autonomous facility surveillance and hazard detection, as well as opening doors to improvements on the factory floor we likely can’t even imagine right now.
What were the biggest technical challenges in enabling AI workloads directly on rugged industrial PCs?
As far as technical challenges go, one of the biggest is that AI requires significant compute power. Increasing power generates heat, and that heat needs to be dissipated. In addition, at the industrial edge, where you have higher temperatures, shock, vibration, and limited ability to perform maintenance (such as replacing fans), that creates a number of design and operational challenges. Emerson’s IPCs were already a big step in that direction, and we couldn’t afford to let the addition of physical AI reduce those capabilities. Fortunately, the SiMA.ai card is built with ruggedization that complements the already robust Emerson IPCs, making it possible to increase compute without reducing performance or reliability.
It is also important to remember that edge machines are expected to run for a very long time. When we considered ways to add additional AI processing power, we couldn’t risk adding components that would shorten the lifecycle of equipment, forcing users to reevaluate new systems and possibly even requiring them to re-qualify those solutions. The industrial nature of SiMa.ai’s MLSoC gives it a similarly long lifecycle to Emerson IPCs so users can standardize equipment with confidence.
How does edge AI improve operational reliability compared to cloud-dependent AI systems?
In an industrial setting, cloud data is becoming increasingly essential. However, getting data to the cloud and back can be a significant challenge. Typically, facilities will need some sort of wireless connectivity, and with machines moving in and out, the quality of wireless links can fluctuate quite a bit. Moreover, the amount of data users will need to send, especially as they inch closer to closing the loop, will be quite high. For less time-sensitive uses, such as historization or predictive maintenance strategies, the added latency of using the cloud isn’t an issue. But as more organizations adopt real-time processing using AI, they need a way to move time-critical data to edge solutions without adding the latency of cloud computing.
Any disruption—physical link availability, cloud outages, service provider disruptions—are all risk factors that could interrupt operations in real-time processing. However, when the team leverages on premise AI on a machine, these disruption factors are mitigated. Moreover, it also provides fault isolation—if there is a failure, one machine stops instead of the whole operation.
We are already seeing powerful AI tools becoming more integrated into processes as a whole. Though these solutions are not typically in the closed loop control yet, there is a potential that as AI evolves in capability, those tools themselves will move from simple advisory roles to more critical components of operations, simultaneously becoming potential bottlenecks if they fail. Bringing many of those tools on site reduces the potential for failure and builds a more robust infrastructure for future deployment.
In addition, the ability to operate locally and avoid sending data back and forth from the cloud can dramatically improve the facility’s cybersecurity posture. Every path the team must open to allow data transmission is a new potential attack vector. Edge-situated AI helps preserve air gapping that helps protect real-time systems.
Can you share a real-world use case where AI-driven edge computing significantly improved productivity or safety?
One recent example is a site manufacturing motherboards that wanted to improve productivity, reduce scrap, and improve quality. The motherboards have two large connectors with many small, through-hole pins. Between the two connectors there are over 500 pins. Traditionally, when an operator installed the connectors, sometimes pins would be bent. If that bent pin went unnoticed, the equipment would go through the soldering process and nobody would know there was a problem until testing, when the item would be scrap. To combat this, the operator would install the connectors, flip the board around, and manually analyze all 500 pins using a microscope—a time-consuming and error-prone process.
The current solution is an IPC with AI software mounted on a cart with a digital camera. The operator can roll the cart to a workstation. The cart camera takes a picture of the motherboard and sends it to the IPC, where the image is run through a model to determine in seconds if the board is assembled properly.
Another example of how these technologies are being used at the edge occurs in oil and gas infrastructure. Cameras at facilities detect leaks, puddles, flare outs, humans in areas they shouldn’t be, elevated temperatures, and more. The AI processes the images in real-time and uses pattern recognition to identify abnormalities, alerting key personnel so they can act before aberrations become accidents. Similar vision inspection could also easily be applied to robots working in isolated areas—clean rooms, for example—to identify issues with equipment performance without needing in-person inspections that interrupt continuous operation or risk injury or contamination.
One key value of using IPCs for tasks like these is that the IPCs run for a long time without failure. For any industry, the long lifecycle is critical for reliability, quality and testing. Finished goods tend to meet specifications when variations are eliminated or reduced. If the IPC is replaced every 2-3 years, that introduces risk and cost. In contrast, more advanced Emerson IPCs should last nearly 10 years.
Cybersecurity is critical in industrial environments. How does your AI platform ensure secure and air-gapped operations?
The best cybersecurity is built using layered defense so if one layer is penetrated, others still provide critical protection. Air-gapped operation is only one of those many layers. With air-gapped operation, bad actors theoretically cannot attack operations in real time, but there are still attack vectors that exist. Perhaps a USB drive or other attack vector delivers a negative payload instead.
This is a reason why IPCs with an integrated IIoT application enablement platform like Emerson’s PACEdge™ bring a great deal of value. The software provides out-of-the-box application-level security, creating another layer. PACEdge also brings a hardened operating system, closed ports, dockerized operation to isolate applications from each other, and more. These tools combine with air gapping to provide the most effective security posture possible.
What role do you think AI-enabled industrial PCs will play in achieving autonomous operations?
Autonomous operation is typically being deployed first in remote locations where there are few or no people, as those are the locations where it typically has the greatest value. In those remote locations, IPCs with physical AI are going to have significant impact in the short term. AI enablement will only strengthen these implementations and help them move more quickly from semi-autonomy to full autonomy as technology allows.
It’s also important to recognize what we’ve learned from the increasing autonomy we’re already seeing in operations. For years, organizations have implemented reliability technologies to unlock predictive maintenance capabilities that are already delivering autonomy and freeing personnel up for more valuable tasks. What we’ve learned from these advances is that autonomy requires massive amounts of data. The closer we can deploy AI tools to where the data is being generated, the better insights we will have—and those insights can be sent up to the cloud when necessary or deployed right at the edge wherever possible.
We’re also seeing massive workforce turnover and with that turnover comes knowledge loss. As that knowledge disappears, organizations need to find a way to embed it within the automation itself. Edge deployments of AI enable that capability. In fact, the next step will be enabling LLMs on the edge. When those systems are deployed, instead of just teaching the AI tools a reflex, the systems will be more intelligent. If things go wrong, the systems will be empowered to make small adjustments and continue to work or take other steps as they adapt. That is the beginning of true autonomy. Then, as those small steps are increasingly successful, we will begin to expand that concept, embedding autonomy in each area of operation. Autonomy will roll out in a process-by-process and segment-by-segment model, creating a layer of autonomy on top of everything. IPCs are at the center of that autonomous operation on a distributed level.
What skills and mindset should future engineers and industrial leaders develop to succeed in the era of Physical AI?
People tend to look at AI as a fun toy, but not particularly useful for their own industrial applications. Because AI appears to not be repeatable, reliable, or guaranteed, it can easily seem as though it won’t ever be a useful tool for real-time operations. But the same has been felt about many previous technologies that are commonplace today.
The first thing people can do is to start experimenting and familiarizing themselves with AI technology. Organizations should encourage their teams to experiment with and learn the technology and not be afraid to perform some proof of concept. Otherwise, it will be hard to embrace the technology when it moves from an edge case to the norm—which is the point at which an organization is already behind the competitive advantage curve. Working with AI requires a different mindset. Practicing those strategies today, and learning where to set the boundaries and how to interact with these new tools can deliver dividends in the years ahead.
In addition, CIOs and CTOs already understand how these tools will tighten supply chains, reduce wastage, improve quality, and limit waste and returns, ultimately increasing revenue. The mandates to implement these technologies are coming. It doesn’t pay to wait until it becomes non-optional. Acting today and staying ahead of the curve lets teams start making the journey more gradually. The needs and business problems already exist, so starting today to catalog those needs and begin pilot projects to see which tools can help can be a critical differentiator in the years ahead.
About the respondents
Gene Juknevicious is senior solution architect for Emerson, recognized for shaping next-generation communication platforms and architectures, with a strong focus on industrial computing and edge-based solutions.
With deep expertise in edge computing, control networks, machine learning and modern computing architectures, Gene is trusted for staying at the forefront of industry evolution and translating complex technologies into scalable, market-ready solutions.
Gene holds a Master of Science in Electrical Engineering from Stanford University and a Bachelor of Science in Electrical Engineering from San José State University.
Manish Sharma is global marketing and business development leader for Emerson’s Controls and Software business and is responsible for its portfolio targeting energy transition across industries like power, renewables, hydrogen, hydrocarbons, semiconductor, and water. He has 25 years of experience in marketing, product management and control systems and R&D. Manish holds a B.E. in Marine Engineering from Marine Engineering and Research Institute (MERI), India and an MBA from IIM, Ahmedabad, India.
