Will “boring” robots win in the real world? Why the robotic form should follow the workflow

Robots have captured our imagination. We’ve seen humanoids balance on one leg, do backward somersaults, fold laundry, and move easily across difficult terrain. All behaviors that only a few years ago seemed like science fiction. These viral examples are important because they show that embedded artificial intelligence has reached a new technical stage.

At the same time, the window on what qualifies as a “humanoid robot” appears to be shifting. Robot companies are adding wheels, more sensors, new manipulators, hats and screens, while removing eyes and heads from their humanoid robot designs.

As an engineering ecosystem that works directly alongside humanoid developers to solve core hardware bottlenecks, we see a recurring theme. The industry often gets caught up in the “robot-forward” mentality. We first build an extraordinary physical platform, then we scour industrial sites looking for a task to perform.

If we want to move physical AI from social media feeds to a sustained commercial level, we need to reverse our current approach. The industry needs a design method that works backwards from the workflow. Start with a task that has economic value. Consider whether a robot can perform this task better than a human.

Then, go back to find out exactly what hardware, sensing capability, and form factor are actually needed to accomplish the task.

The sensor stack is born in the business, not in the data sheet

If you start by designing a platform, you will likely end up with a compromised hardware stack. The specific workflow determines the technical requirements of the machine.

Let’s take as an example a bipedal humanoid robot that is asked to carry out two different industrial work processes. The first involves placing a pin on an automotive assembly line, the second involves packing a pallet in a logistics center.

The operation of inserting a pin connector typically involves a gap of 20 µm and a mating force of 5 N. Therefore, a contact force resolution of less than 0.1 N and a control bandwidth fast enough to detect a jam before a sensitive pin bends is required.

However, when using the same arm for packing loose pallets, almost no high-frequency microforce resolution is needed. Instead, high load capacity, structural rigidity and the ability to withstand 1 million cycles without basic mechanical drift are required.

If you are building a general purpose platform, a single “middle of the road” force sensor can result in a system that may make precise pin insertion difficult. The system may also lack the durability needed for continuous heavy lifting.

The workflow also determines the error budget. This is the one factor that transforms force sensing from a cost item into something that provides a demonstrable return on investment.

Attempts to improve force sensing often fail. For example, our tests have shown that using a specific calculation while maintaining the contact force at ±0.2 N can reduce connector waste by between 1.2% and 0.5%. This can translate into an estimated annual savings of approximately $450,000 per assembly line. This type of economic argument is only possible if you start with the task at hand.

This is exactly why the selection of critical components, such as those based on strain detection, must occur at the architectural design stage rather than being treated as an afterthought during bill of material (BOM) preparation.

Adjust the robot instead of adjusting the structure

The tendency to focus on the task first raises the main question about the form factor. Is a specialized form factor the best choice? When is a humanoid project strictly necessary?

Operating environments can be divided into two types: greenfield and brownfield. The real question is not what the robot should look like, but whether it is cheaper to modify the robot or modify the system.

With greenfield sites it is possible to use specialized robots. With brownfield sites, such as an existing refinery or chemical plant, the robot must adapt to spaces created for humans.

In greenfield construction, such as Equinor’s Northern Lights facility, the operator can design the layout from the start and build the facility based on the task. Therefore, specialized, non-humanoid forms almost always prove better in terms of unit economics, speed, and energy efficiency. Robots with legs and wheels are more reliable for the types of tasks they perform.

However, in most cases, industrial activities take place in brownfield sites, where the only practical option is to adapt the existing platform. These sites have narrow corridors, stairs, floor grates and doors designed with the human body in mind. In this situation, a human-friendly form factor, whether involving legs, two arms, or a humanoid shape, becomes a great advantage. It allows equipment to operate in environments designed for people without spending millions of dollars retrofitting facilities.

The real test for a humanoid project is whether, if you had to start over, you would still choose a humanoid. When human compatibility is needed, the basic physical requirements remain the same: foot contact, joint torque and wrist strength.

Robots may have different shapes, but they have the same sensory physics. Deformation-based sensing endures even as the form factor changes, offering critical ground truth measurement regardless of how your physical platform develops.

The depth of deployment exceeds the breadth of the demo

Flashy viral videos show what’s technically possible. The real proof, however, is not the demonstration itself, but the second purchase order. Only through repeated orders can business usefulness be demonstrated.

While demonstrations with calibrated units under controlled climate conditions demonstrate that a robot can operate once, fleet-wide deployment requires the robot to survive at least 10,000 cycles per shift under conditions of constant thermal variation and with near-zero tolerance for drift.

Most physical AI projects fail because the engineering effort instead turns to broad, general promises before addressing specific operational facts. In real-world situations, implementations fail because of drift, sensor fatigue, and recalibration difficulties, not because of reasoning flaws.

Editor’s Note: Humanoids and physical AI are among the topics covered during the RoboBusiness 2026 session, which will be held on October 20 and 21 in Santa Clara, California. Register now to participate.

The physical balance of AI

Integrating AI into physical hardware requires a disciplined balance between power consumption, thermal dissipation, mechanical wear and security limitations. With software you grow by making copies; with hardware, you have to buy it, build it, and wait.

Teams that take a software approach often fail to account for this physical balance. Beyond basic thermal or electrical limits, gear reductions of nonlinear actuators introduce complex behaviors that are notoriously difficult to predict and simulate.

True backward driveability solves part of this problem by allowing the robot’s actuator or joint to fully reflect any physical impacts from the environment to the main control architecture.

Whatever the hardware bottleneck, whether it’s a thermal limitation in the joint actuators or the amount of power drawn under peak load, each constraint boils down to the same need: knowing exactly what force is present under actual load conditions, at real temperatures, over a period of years. This is the information provided by strain-based sensing, which is why it should be included in the architecture and not simply listed in the bill of materials.

The true parameters of commercial feasibility

To evaluate the commercial viability of physical AI systems, we should go beyond simple marketing videos. Instead, we should focus on the metrics that finance teams actually care about.

The most important and essential metric is the mean time between interventions (MTBI). This may be considered the most authentic measure used in the industry, but it is the least often published. If a human operator has to intervene every 20 minutes to unlock a gripper or reset a displaced sensor, high autonomy values ​​​​have little meaning.

We must give equal importance to the actual cost per unit of work, whether calculated per sample, per meter traveled or per inspection completed. This figure ultimately determines whether a fleet of robots will replace or augment existing manpower.

The operational value also depends on the redistribution time. Implementation time shows whether a “generic” platform can quickly adapt to a new business or whether reconfiguration requires weeks of custom design. Ultimately, the number of safety incidents per hour of operation will determine whether the robot will ever emerge from its cage to operate freely among human workers.

A viral video shows a robot capable of performing a certain task once. However, a sustainable business needs the robot to perform the same task at least 10,000 times in a row, without any intervention, drift or failure.

The main conclusion is simple: start with workflow, integrate reliable sensing into the core architecture, and focus on operational metrics so that both specialized machines and humanoid platforms can realize their potential in the physical world.

About the author

Ron Zukerman is director of innovation at Vishay Precision Group Inc. (VPG). The company is a leader in precision measurement and sensing technologies.

VPG said its sensors, weighing solutions and measurement systems optimize and improve the performance of customers’ products across a broad range of markets to make the world safer, smarter and more productive. To find out more visit the company on www.vpgsensors.com and follow him LinkedIn.