Investment in robotics and physical AI is rising quickly, but money alone does not make an autonomous factory possible. The harder problem may be the information available inside the factory. Industrial robots need data that connects what happened during a process to the quality of the finished result. In many facilities, that connection is incomplete, scattered across systems, or never captured in a form machines can use.

That is the argument made by Mike Kenworthy, chief technology officer at VulcanForms, in an analysis of the manufacturing data gap. Robotics and physical AI companies have raised nearly $48 billion so far in 2026, according to the article. Yet the transition from controlled robotics demonstrations to autonomous systems working across real production floors depends on detailed, contextual operating data that many plants were not built to provide.

Robots need more than machine readings

Manufacturing sites already generate large volumes of operational data. The issue is not simply a lack of sensors or records. Much of the information was collected for monitoring, reporting, or isolated process control rather than for training systems that must make and evaluate decisions autonomously.

Consider welding. A robot intended to improve its own welding technique would need to understand how changes in angle, travel speed, workpiece position, and other parameters affect the completed weld. Skilled technicians can develop that understanding through years of direct experience. But when their adjustments and observations are not recorded consistently, that practical knowledge does not become usable training material for an autonomous system.

The same pattern can appear in other factory processes. Operators may make adjustments based on tacit knowledge, changing conditions, or visual assessment. If those decisions are not linked to measurable outcomes, a machine-learning system lacks the evidence needed to determine whether a change improved quality, reduced defects, or created a new risk.

The problem of disconnected systems

Factories often store relevant information in separate systems. A production machine may log its operating settings while a quality-management system records inspection results elsewhere. The two data streams can each be useful on their own, but they are far more valuable to physical AI when they can be connected.

Without that link, an autonomous robot may know that a machine used a particular setting without knowing whether the part made under that setting met the required quality standard. It may see an inspection failure without having a reliable record of the process conditions that preceded it. The missing relationship between action and outcome makes adaptation difficult.

The article cites the Manufacturing Leadership Council’s 2026 Smart Factories and Digital Production Survey, in which 37% of surveyed manufacturers identified a lack of data interoperability as a major obstacle to their smart-factory strategies. That was up from 22% in 2025. The increase suggests that companies are recognizing the issue more clearly as their digital ambitions expand.

Interoperability, however, is only part of the challenge. Connecting systems can make information easier to access, but it does not automatically establish whether the data is sufficiently detailed, reliable, and tied to actual product quality. Physical AI needs context as well as connectivity.

From data collection to decision-quality data

For an industrial robot to make reliable adjustments, it must have records that reflect real production conditions. That means understanding which variables matter for a given task, how they were measured, and whether the resulting information can be associated with an outcome that matters to the manufacturer.

A useful dataset for physical AI therefore goes beyond a collection of machine logs. It should help answer practical questions: What action was taken? Under what conditions? What happened to the part or process afterward? And was that outcome acceptable according to a measurable quality standard?

The article warns that simply collecting more information will not solve the problem. Additional data that remains fragmented, poorly contextualized, or detached from finished-part quality may add volume without improving an autonomous system’s ability to learn. Manufacturers must instead evaluate the information their robotic systems will depend on and determine whether it supports reliable adjustments over time.

Why the gap matters for deployment

Physical AI is often discussed as a next stage for robots: systems that can work in dynamic settings, react to variation, and improve through experience. Manufacturing environments are a demanding test of that promise. Parts, tools, materials, and operating conditions can change, while quality requirements may remain strict.

In that setting, an AI system that cannot connect its behavior to verified outcomes may struggle to adapt safely or effectively. It may be capable of sophisticated perception or motion, yet still lack the evidence to know whether a process change is helping. The result could limit the efficiency and quality gains manufacturers expect from autonomous machines.

The data gap also highlights that deploying physical AI is an organizational challenge as well as a robotics challenge. Capturing technician knowledge, integrating production and quality records, and defining meaningful feedback loops require work across operations, engineering, and data systems. The factory has to become a learning environment before a robot can learn reliably within it.

Preparing the factory for autonomous learning

The source frames preparation for physical AI as more than a technology-purchasing exercise. Manufacturers need to identify the process information that autonomous systems will rely on, examine whether it accurately represents production, and ensure that it can be related to measurable quality outcomes.

That focus moves the discussion away from a simple question of whether robots are advanced enough. In many cases, the constraint may be whether the industrial environment supplies the structured experience those robots need. Autonomous machines can only learn from what a factory makes visible.

As robotics investment continues, manufacturers that close the distance between process decisions and quality feedback may be better positioned to put physical AI to useful work. The immediate task is not merely to gather more data. It is to make production data coherent enough that a machine can learn which actions lead to better results.

This article is based on reporting by The Robot Report. Read the original article.

Originally published on therobotreport.com