At TU Wien, researchers have developed a new approach to teaching robots how to perform tedious tasks such as cleaning. Instead of relying on fixed rules and predefined mathematical formulas, humans demonstrate the task to the robot multiple times using a specially prepared cleaning sponge. By observing humans, the robot learns how to clean and can apply this knowledge to different shapes and surfaces. This innovative learning process has been showcased in a study published at IROS 2024 in Abu Dhabi.
In addition to cleaning, the technology developed at TU Wien can also be applied to other surface treatment tasks such as sanding, polishing, painting, and applying adhesives. Teaching a robot to perform these tasks involves much more than capturing the geometric shape of the object with cameras. The robot must also learn the appropriate movement, speed, angle, and force required for each part of the surface. This is achieved through a combination of statistical data processing and training a neural network to learn predefined movement elements.
The research team at TU Wien has developed an innovative data processing strategy that enables the robot to learn from a few demonstrations and effectively control the robot arm to clean the surface optimally. This learning algorithm allows the robot to clean the entire sink or other objects with complex surfaces after being trained on just a single edge of the sink. The robot can adjust its technique based on the shape of the surface, applying different amounts of force depending on the curvature of the area being cleaned.
The vision for this technology extends beyond just cleaning sinks; it could be used for various processes in different industries. The robot could be placed on a mobile platform to serve as a helpful assistant in workshops, learning from other robots and sharing knowledge through a concept known as ‘federated learning.’ This allows robots to share basic principles and improve their capabilities collectively while keeping private data, such as specific workpiece shapes, confidential.
The flexibility and effectiveness of the sink-cleaning robot technology have been demonstrated through numerous tests at TU Wien, garnering international recognition at IROS 2024. The research was awarded the ‘Best Application Paper Award’ at the conference, highlighting it as one of the top innovations of the year. This breakthrough in robotics paves the way for a future where intelligent robots can learn and adapt to a wide range of tasks, making them valuable assets in various settings.