Researchers develop self-learning robot that can clean washbasins like humans

On November 10, researchers at the Vienna University of Technology (TU Wien) developed a self-learning robot that can imitate humans to complete simple tasks, such as cleaning a washbasin.

This seemingly simple task is a huge challenge for robots. Traditional robot programming requires precise mathematical modeling of the complex curves of the washbasin, which is a time-consuming and complex task. The research team simplified the process by combining human demonstrations and tactile data to train the robot to imitate human movements.

This learning method is not only suitable for household cleaning, but can also be applied to a variety of tasks in industrial production, such as polishing, painting, sanding and gluing.

"It is relatively easy to capture the geometry of the washbasin with a camera," says Professor Andreas Kugi from the Institute of Automation and Control at TU Vienna. "But the key is to teach the robot how to choose the right movement for each surface. How fast should the movement be? What is the right angle? What is the right force?"

IT Home noticed that this way of learning is similar to the process of humans learning new skills, especially skills that require hands-on operation.

“In the workshop, a master will instruct an apprentice and say, ‘You need to press harder on that narrow edge,’” says Christian Hartl-Nesic, head of the industrial robotics group in Andreas Kugi’s team. “We wanted to find a way for the robot to learn in a similar way.”

To this end, the research team developed a special cleaning tool - a sensor-impregnated sponge. The human "teacher" used this sponge to repeatedly clean the front edge of the sink using force sensors and tracking markers.

“We generate a lot of data with just a few demonstrations, which we then process to help the robot understand the correct way to clean,” explains Christian Hartl-Nesic.

The research team then used an innovative data processing strategy to enable the robot to effectively clean an entire sink or other object with a complex surface, even if it had only been shown how to clean one edge of the sink.

"The robot learns to adapt the way it grips the sponge to the shape of the surface and to apply different forces on tightly curved areas than on flat surfaces," explains Christoph Unger, a doctoral student in the Industrial Robotics Group.

In the future, robots could be mounted on mobile platforms, making them useful assistants in any shop floor environment. And just like humans, they could pass on their experience to other robots.