Workflow First: Why Robot Design Must Start With the Task, Not the Platform

Humanoid robots are evolving fast — gaining wheels, sensors, manipulators, hats, and screens. But according to Ron Zukerman, innovation director at Vishay Precision Group Inc, the industry's fundamental problem isn't capability. It's approach.

"If we want to move physical AI from social media feeds to a sustained commercial level, we must reverse our current approach. The industry needs a design method that works backwards from the workflow."

Two tasks, two entirely different machines

The workflow determines the machine's engineering requirements, not the platform design. Zukerman illustrates this with a bipedal humanoid robot asked to perform two industrial tasks: inserting a pin on an automotive assembly line, and packing a pallet in a logistics center.

The pin insertion task involves a 20 µm clearance and 5 N mating force. That demands contact-force resolution below 0.1 N and control bandwidth fast enough to detect jamming before a pin bends. Pallet packing requires almost none of that micro-force resolution — instead calling for heavy payload capacity, structural rigidity, and the ability to endure 1 million cycles without mechanical drift.

The takeaway for engineers: using a single force sensor for diverse tasks can lead to compromised performance. A "middle-of-the-road" sensor makes precise insertion difficult while lacking the durability for continuous heavy lifting.

The ROI case for task-driven sensing

Workflow-based design delivers measurable economic benefits. Maintaining contact force at ±0.2 N reduced connector scrap from 1.2% to 0.5% — translating into estimated annual savings of approximately $450,000 per assembly line. Strain-based sensing provides the critical force information under actual loading conditions that makes this possible, which is why it belongs in the core architecture rather than the bill of materials.

Greenfield vs. brownfield: the form factor question

Greenfield sites, such as Equinor's Northern Lights facility, allow operators to design layouts around the task — favoring specialized robots. Brownfield sites, with their narrow aisles, stairs, and doorways built for people, require adaptable, human-suitable form factors. Humanoid designs prove advantageous where modifying the facility would cost more than adapting the robot.

"The true test for a humanoid design is whether, if you were starting again, you would still go with a humanoid."

Whatever the shape, the sensory physics stays constant: foot contact, joint torque, and wrist force. Strain-based sensing endures across form factors.

Metrics that matter more than demos

The true test for robots is not demonstrations but repeat orders and sustained performance. Demos prove a robot works once; fleet deployment demands it survive at least 10,000 cycles per shift amid thermal variation with near-zero drift tolerance.

"A viral video shows a robot capable of carrying out a given task once. However, a sustainable business needs the robot to carry out that same task at least 10,000 times in a row, without any intervention, drift, or failure."

"The most important and essential metric is mean time between interventions (MTBI)."

If a human must intervene every 20 minutes to unjam a gripper, autonomy figures mean little. Physical AI systems must also balance power consumption, thermal dissipation, mechanical wear, and safety limitations — a discipline software-first teams often miss. True backdrivability helps, allowing robots to reflect environmental impacts back to the control architecture. Operational value ultimately hinges on redeployment time and safety incidents per operating hour.

The conclusion is direct: start with the workflow, build reliable sensing into the architecture, and measure what finance teams actually care about.