
Robotics and physical AI companies have raised nearly $48 billion so far in 2026, but the rollout of fully autonomous machines in industrial settings isn’t a given. For robots to step beyond controlled environments onto the manufacturing floor, they need real-world data to learn from.
Most production facilities aren’t set up to provide it.
Consider the welding process. For a robot to improve its technique autonomously, it needs to understand how changes in welding angle, travel speed, workpiece position, and other key parameters affect the finished result. Experienced technicians develop this understanding through years of hands-on experience, but much of that knowledge isn’t captured in a form robotics partners can use to train autonomous systems.
Similar gaps exist in other manufacturing processes where technicians rely on experience to make adjustments that aren’t consistently documented.
Manufacturers looking to deploy physical AI need to close this data gap. Without information connecting process decisions to production outcomes, even sophisticated robotic systems may struggle to adapt to changing conditions.
These limitations could prevent manufacturers from achieving the efficiency and quality improvements they expect from physical AI.
Why industrial data remains a barrier to physical AI
Manufacturers already collect extensive operational data, but much of it was never structured for autonomous learning. Sometimes it is collected and never even reviewed at all!
Relevant contextual information often sits in separate systems. A machine may record operating parameters while a quality management system captures inspection results, making it difficult to connect process decisions with finished-part quality.
Manufacturing leaders are increasingly aware that these silos are an obstacle. According to the Manufacturing Leadership Council’s 2026 Smart Factories and Digital Production Survey, 37% of surveyed manufacturers cited a lack of data interoperability as a major roadblock to implementing their smart factory strategies, up from 22% in 2025.
However, improving interoperability alone won’t get robots the data they need to learn. Manufacturers also need to ensure that production records can be tied to measurable quality outcomes.
3 ways manufacturers can prepare for physical AI
Preparing a factory for physical AI requires more than collecting additional data or connecting disparate systems. Manufacturers need to understand what information their robotic systems will rely on, how that data reflects real production conditions and whether it can help systems make reliable adjustments over time.
Addressing these questions early, in collaboration with robotics partners, can help manufacturers identify gaps in their existing infrastructure and focus investments where they’ll have the greatest impact.
Here are three areas to prioritize before deployment.
1. Capture data that reflects real production conditions
Begin by identifying the production conditions your robotic systems need to recognize and respond to. Focus on capturing the complex interactions between the process parameters and quality outcomes to show how planned and unplanned changes in materials or equipment performance affect the finished product.
Consider metal additive manufacturing, where factors like powder (or wire) characteristics, processing temperatures, and environmental conditions can influence a component’s final properties and characteristic defect structure. Connecting these variables with inspection results gives your robotics partners a clearer picture of which conditions contribute to defects or variations in quality.
You can then use that information to train systems to recognize changing production conditions and develop more realistic simulations for testing their responses.
2. Connect data across the entire production process
The data you capture won’t have much value if it remains isolated within individual machines or production stages. The next step is to create a digital thread that connects production conditions with inspection results and finished-part quality.
I’ve seen this challenge firsthand in advanced manufacturing. AI-based image recognition is becoming table stakes for assessing quality within metal additive manufacturing printers, while downstream equipment such as heat-treatment furnaces may be far more difficult to integrate into the same system. Lior Susan of Eclipse rightly says: “If your physical asset isn’t generating real-time operational data, it’s a dead asset.”
The disconnect creates gaps in the production history of a component, making it harder to understand how subsequent processing affects its final properties.
To close those gaps, you need to connect equipment and quality management systems that weren’t designed to communicate. Start by identifying the data needed for a specific robotic application, then work with technology partners to establish consistent part identifiers and link process records to inspection results in a true digital thread.
By starting with a targeted application, you can build the data connections to support physical AI without having to overhaul your entire production infrastructure at once.
3. Build feedback loops around quality and repeatability
Finally, establish how autonomous systems will be evaluated and when human oversight is required. Work with your robotics partners to identify which process adjustments can be made automatically and how the system will determine whether those changes improved production outcomes.
In medical device manufacturing, for example, regulatory requirements often mandate validation of production processes to ensure consistent product quality. Changes to those processes may require additional testing and revalidation, limiting how freely AI systems can modify production parameters.
Before authorizing autonomous adjustments, work with your robotics partners to define acceptable operating limits and establish how the system will verify results.
Also develop clear criteria for when an unexpected outcome requires engineering review. A system that can modify operating parameters but can’t evaluate the resulting quality is not ready to make those decisions independently.
Preparing manufacturing data for industrial autonomy
As you evaluate data infrastructure investments, determine whether they will support your physical AI goals. Before adding more sensors or connecting another piece of equipment, have a plan for how the resulting information will help explain how a process adjustment affects product quality.
Where that connection is missing, you may get more value from linking existing process records to inspection results than from collecting additional machine data.
Industrial autonomy won’t depend on how much data manufacturers collect, but on whether their robotic systems are actually leveraging that data to make reliable decisions on the factory floor.
About the author
Michael Kenworthy is chief technology officer at VulcanForms, bringing more than a decade of experience in engineering, product development and advanced manufacturing. Previously, he served as VP of engineering at Relativity Space, where he helped lead the development of 3D-printed rockets, and chief product officer at Seurat Technologies, where he oversaw product strategy. He holds degrees from MIT and the University of Wisconsin-Madison.





Tell Us What You Think!