By Gleb Tsipursky, PhD, CEO, Disaster Avoidance Experts
A new industrial robot can meet its cycle-time target and still fail its most important workplace test: what happens when an experienced operator believes the machine is wrong?
That question deserves attention as manufacturers accelerate automation. The International Federation of Robotics reported in September that the global stock of industrial robots exceeded five million units in 2025, with more than 600,000 new installations that year.
Increased capacity is real. So is the need to know how employees can challenge the system before a mistake spreads across a shift.
The latest performance improvements make the issue more urgent. As Robotics & Automation News reported on October 20, ABB’s new IRB 1700 models promise up to 20 percent shorter cycle times, with precision and energy-efficiency gains.
Faster cycles help a business meet orders. They can also reduce the time in which a worker identifies and explains an unexpected result. Speed magnifies the value of a clear escalation process.
In a recent opinion on simulation before deployment, BlueBotics CEO Nicola Tomatis made a useful case for testing traffic, charging, routes and fleet capacity while changes remain inexpensive. There is another variable worth testing before a system goes live: the human response to a disagreement.
Imagine a machine-vision system marks a batch of parts as defective. A veteran operator suspects residue on the camera lens, not a production defect. The automated workflow is ready to divert the batch.
Who can challenge that classification? Can they pause the quality decision without compromising machine safety? Which supervisor receives the evidence? Does the next shift know what happened? This is a hypothetical example, but the questions are ordinary operational ones.
Safety procedures must remain separate and paramount. OSHA’s robotics guidance notes that many accidents occur during non-routine operations such as testing, setup, maintenance and adjustment.
A disagreement protocol cannot replace physical safeguarding, lockout/tagout, emergency stops or task-based risk assessments. Workers should never have to enter a robot’s hazardous operating area to contest an automated recommendation.
Instead, give operators a safe route to challenge the decision itself. I would include three rehearsals in commissioning. First, introduce a plausible false positive, such as a harmless part flagged by inspection software.
The operator should be able to flag the error, preserve the evidence and get a timely decision from the responsible quality lead.
Second, present a conflicting instruction. The production planner wants throughput, while a technician has observed a pattern that needs investigation.
A clear decision-rights map should identify who can order a pause, who investigates, and who is accountable for restarting work. Nobody should have to infer authority from a job title in the middle of a shift.
Third, test the changeover between people. The overnight operator discovers an intermittent fault; the morning crew inherits a clean dashboard because the alert disappeared after a reset.
The system should retain an exception record with the concern, the evidence, the response, and its status. If a report disappears with the shift, an apparently efficient factory has lost important information.
This is more than a theoretical governance exercise. NIST’s 2026 roadmap for AI and machine learning in smart manufacturing identifies reliable, explainable operation and the integration of complex industrial systems as central challenges.
Its separate AI for Manufacturing research initiative explicitly includes metrics for human-AI teaming and methods to evaluate operator understanding of AI solutions. Those are practical priorities for a plant manager buying into robotics today.
An effective commissioning scorecard should therefore report more than line speed, accuracy and uptime. Add the time to register an operator challenge, the time to reach a qualified decision-maker, the share of challenges resolved with documented evidence, and the number of recurring faults discovered across shifts.
An exception that gets raised and resolved may be a sign of a healthy system, not an inconvenient failure.
The numbers also offer a better way to evaluate training. Ask an operator to explain what the robot can determine, what data it cannot see, and what they should do when observations conflict.
Give them a realistic scenario to practice. Watching a vendor demo is useful orientation; handling a disagreement safely and correctly is stronger evidence of readiness.
There is a cost objection. A rehearsal takes time, and managers want the equipment producing as soon as it is installed. Yet OSHA’s robotics standards overview points manufacturers toward task-based risk assessment and safeguarding guidance precisely because physical systems require deliberate preparation.
Testing decision handoffs alongside safety measures belongs in that preparation, even though the two serve different purposes.
A factory can buy ever faster robotics. It cannot buy away the need for judgment on a noisy, changing production floor. Before signing off on commissioning, ask one operator to disagree with the robot while the whole team watches the process. If the response is uncertain, the workflow is not ready for a production shift.
About the author: Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook
