In just a few seconds

TU Vienna robot learns to clean washbasins

(Bild: TU Wien)

Who wouldn't wish for this when faced with a dirty washbasin - a robot system that learns in seconds to take over the unglamorous task without causing damage to the porcelain surface. Viennese researchers have now realized such a robotic AI system. However, the approach promises far more than just a flexible cleaning robot.

The work of researchers from the Institute of Automation and Control Engineering (ACIN) at TU Wien was selected for a "Best Application Paper Award" from around 3,500 papers submitted to the "IROS" robotics conference recently held in Abu Dhabi and was therefore ranked among the most important innovations of the year by peers. Andreas Kugi and Christian Hartl-Nesic from ACIN explained that this shows that "useful things" are being done in Vienna in this "highly dynamic scientific field worldwide".

Learning with a sponge
In this specific case, the AI system learns more or less directly from humans. The vehicle for this is an "instrumented tool" - in this case a standard household sponge that the scientists have simply perforated and fitted with force-torque sensors. This records in great detail how much force is applied when handling the sponge, at which point and at which curvature of the workpiece - in this case the porcelain basin.

Watching is enough
The movements of the leading hand are also recorded precisely in 3D using an optical tracking system. In combination, this provides hundreds to thousands of measurements per second. This very high resolution also explains why the downstream data processing system, which is based on artificial intelligence or machine learning, can make a pretty good guess about the requirements after just a few seconds of "watching". "We can extract a very good understanding of the process," as Hartl-Nesic put it.

Behind the "learning" is a mathematical-statistical approach with which similarities in the movements can be broken down. In this way, what is shown can be technically translated in the form of very few parameters and ultimately imitated, explained the scientist: "For each new edge, the system can use the statistical description to calculate a meaningful tool movement."

The demonstrations showed that it can replicate the sink task after only a minimal demonstration - a person cleaned the front edge of the sink for just a few seconds by simply wiping it - and can also apply it to new situations, such as other curvatures. In other words, cleaning the rest of the geometrically quite challenging object will then take care of itself. In future, the robot will also check its success after the process has been carried out so that it can rework if necessary.

The cleaning robot at work

(Bild: TU Wien)

Keep waiting
However, this assistant will not find its way into households any time soon: the high-tech robotic arm and the advanced system developed in Vienna are not available to buy off the shelf in this form. Kugi: "There is still a lot of basic research to be done here", but the ingredients are there.

The researchers' vision is of "robots as assistants". Kugi is convinced that this is particularly important for Austria as a country with many small and medium-sized manufacturing companies in the craft sector in the medium term - also in view of the increasing shortage of skilled workers. Such systems would "not replace people, but provide massive support".

Area of application: More or less everywhere where surfaces of all kinds have to be sanded, polished, painted, adhesives applied or parts joined together.

Now is the time for companies to think about what their production will look like in ten to 15 years' time. "Intelligent assistance systems with semi-autonomous functions" are one part of this. It would also be conceivable to gradually and jointly improve such approaches across several companies and applications: "Private data - for example about the specific shape of a particular workpiece - would remain private, but important basic principles learned would be exchanged in order to further improve the capabilities of all robots," says TU Wien.