Automation without perfection: How intelligent robots can handle HMLV manufacturing

A factory that cannot be captured in a spreadsheet

The traditional approach to automation is usually based on a relatively simple assumption: if a person repeatedly performs the same operation, a robot should be able to replace them. In electronics manufacturing, however, automation projects are rarely that simple. Piotr Owczarek, who has worked in electronics manufacturing for around 25 years, noted that he had repeatedly witnessed the same scenario:

“Almost every time — in fact, every time — the result was the same: I replaced one operator with one robot and an engineer,” he said.

At the business-planning stage, the investment often looked attractive. Spreadsheet calculations indicated lower labour costs, higher production rates and a relatively short return on investment. The situation changed once the system was deployed on the factory floor. Production was not as regular as the model had assumed. Robots required supervision, adjustments and frequent technical intervention. The operator disappeared from the workstation, but an engineer appeared next to the machine. The automation worked, but its costs and organisational requirements were considerably higher than expected.

According to Owczarek, the issue is not simply that robots lack sufficient capabilities. The larger problem is that conventional automation is poorly matched to the conditions in which many European electronics manufacturers operate.

High mix, low volume

European electronics manufacturing frequently follows the high-mix, low-volume model, known as HMLV. This means producing many different types of products in relatively small batches. A factory may spend one day assembling a particular device and the next manufacturing a different PCB variant, a revised product or an entirely new design.

These conditions are not well suited to conventional, rigidly configured automated production lines. When a production run consists of only a few dozen or a few hundred units, it is difficult to justify the cost of preparing the workstation, developing dedicated tooling and programming the robot. Owczarek illustrated the situation by recalling an argument frequently made by manufacturers:

“We produce 100 units. By the time we set up the process, we have already completed the entire year’s production.”

Schedule variability creates an additional challenge. Even perfectly planned production can be disrupted by a delay involving a single component. The sequence of orders and factory-floor operations must then be adjusted to reflect the current availability of materials. A human usually responds to such changes intuitively: they notice a new package, a repositioned container or a different PCB variant and adapt their actions accordingly. A conventional robot is not nearly as flexible. It follows a programmed sequence based on the assumption that all predefined conditions will remain unchanged.

Electronics manufacturing demands high precision

Automation in the electronics industry is further complicated by the small size of the parts being handled. A robot moving cardboard boxes or large mechanical components can tolerate a certain degree of inaccuracy. With PCBs, connectors and small electronic components, even a minimal displacement may prevent the operation from being completed correctly.

The AiRob expert referred to the example of automated testing. In theory, a robot can pick up a PCB, place it in a tester, close the device and then transfer the product to the next stage once the procedure has been completed. A single machine can therefore operate several testing stations. In practice, however, it is enough for the tester to move by a fraction of a millimetre. The PCB may then fail to enter the socket correctly, bringing the entire process to a halt. Human intervention and a correction to the program become necessary.

“The robot is standing there with one or two engineers next to it. It is a very common sight in electronics manufacturing,” Owczarek said.

AiRob develops intelligent robotics systems intended to reduce the need for mechanical positioning. Instead of fastening every piece of equipment to the floor or a workbench and requiring it to remain in exactly the same location, the robot should identify its surroundings and adapt its movements. Owczarek described this concept as “cutting the Gordian knot of robotisation.”

A robot that sees before it moves

Perception is the first stage of intelligent automation.

“The robot has to see. It has to assess what needs to be done. It has to use algorithms to analyse the task in front of it,” speaker explained.

The image is interpreted using algorithms, 3D models and artificial intelligence tools. Only then does the robot make a decision and perform the operation. Its movements are not based solely on a rigidly programmed sequence. The system identifies objects, evaluates their accessibility and selects an appropriate action based on the current situation.

Piotr Owczarek referred to an AiRob deployment involving components being picked from trays. Although the parts are arranged in an orderly manner, the plastic packaging may bend or move. The vision system therefore checks the position of the components every time. The robot does not have to pick them up in a predetermined sequence. It can select the part that is most accessible at that moment, shortening its movement and reducing the risk of a collision.

Working with randomly arranged components is a much more difficult task. The system creates a three-dimensional model of the situation. It then determines which components can be picked up safely, where they should be gripped and whether the movement of the gripper could damage other parts.

Artificial intelligence is particularly useful when the product changes. As Owczarek explained, AiRob uses algorithms trained on standard electronic components. This allows the system to handle new parts based on experience acquired while working with similar objects.

Quality inspection before assembly

Once a component has been picked up, its precise position in the gripper must be established. The component pins are particularly important. The vision system can determine where they are located, whether they are straight and whether their number matches expectations. A correct component can be placed on the PCB or sent to the next operation. A damaged part is rejected.

Importantly, AiRob’s model does not assume that humans will be eliminated from the process entirely. Components rejected by the system can be passed to an operator, who assesses their condition and, when necessary, may straighten the pins.

“With today’s technology, we will not automate 100% of production,” Owczarek said. “We are still a long way from replacing humans entirely.”

He believes that automating around 70–80% of production operations is a more realistic objective. People would continue to handle unusual cases, exceptions and tasks requiring experience or judgement.

The PCB does not have to be perfectly aligned

One of the most illustrative examples presented during the talk involved picking up PCBs that had deliberately been arranged unevenly. The robot recognised their positions in three-dimensional space, identifying the x, y and z coordinates as well as the angles of rotation. It could use reference markers, corners and other distinctive points on the board to locate them.

“It was not programmed for this at all. The decision is made in real time by the algorithms,” the AiRob CEO explained.

The same approach can be applied to testers, workbenches and carts carrying components or finished products. An operator does not need to place a cart with sub-millimetre accuracy. It only needs to be positioned within a designated area. The vision system can then determine its actual location. As a result, the production environment no longer needs to remain perfectly static for the robot to operate.

Modular automation

The second pillar of flexible automation is modularity. Traditional production lines are often designed for a particular customer and product. Such investments can be justified when production volumes are high and stable. However, market conditions can change quickly. Demand may decline, a customer may introduce a new product revision, or a product may prove to be seasonal. A line expected to operate for several years can become underused or obsolete.

A modular robot can be disconnected from one workstation and moved to another. An operator can take over the original process while the robot is assigned to a different task.

Automation is therefore no longer permanently linked to one production process. Instead, the robot becomes a flexible factory resource that can be moved according to current orders, production priorities and component availability.

Artificial intelligence remains inside the factory

Robots equipped with cameras may process highly sensitive information about products and manufacturing processes. Owczarek said AiRob’s first systems were based on cloud computing. The company quickly discovered, however, that manufacturers were reluctant to connect a camera-equipped robot operating on the production floor to an external service.

“No one today will allow a robot with a camera operating on the production floor to be connected to any cloud,” he said.

AiRob’s current systems use a factory server physically installed at the manufacturing site. It can operate entirely within the company’s internal network without requiring a permanent external connection. Updates, additional knowledge and new functions can still be delivered, but in a controlled manner that does not expose production data to an external cloud platform.

Building robots for an imperfect factory

The main idea behind intelligent robotics is to give machines the ability to respond to imperfections and changes in their surroundings. Within a clearly defined range of tasks, the robot should be able to see, analyse and make decisions in a way that resembles the work of a human operator. Generic models can handle the majority of standard PCBs and electronic components. Less common cases can be added to the system gradually.

“When we have generic models that cover 90% of solutions, we will always have enough time to provide additional training for the remaining 10%,” Owczarek said.

According to the AiRob CEO, the combination of machine vision, 3D modelling, artificial intelligence and modular robotics can help automate processes that were previously not considered suitable for automation because of their high variability and low production volumes. A modern automated factory is therefore unlikely to be completely free of people. Robots may take over most repetitive operations, while employees focus on exceptions, supervision, quality control and tasks requiring experience.

For European manufacturers, this flexibility may prove more valuable than maximising the speed of a single production line. Owczarek argued that automating 70–80% of production could help keep electronics manufacturing in Europe. Achieving this objective, however, requires manufacturers to accept that a real factory will never be as predictable or as orderly as the model created in an Excel spreadsheet.

The next Polish edition of Evertiq Expo will take place on October 22, 2026, at PGE Narodowy in Warsaw. Evertiq Expo Kraków will return on June 9, 2027