Watch: Ambitious Robot Learn Show to Clean a Bathroom Sink by Observing

From washing urinals to tidying beaches, the idea of robots helping keep our world cleaner is already becoming a reality. Now, a robot arm has successfully tackled the surprisingly intricate task of washing a sink, demonstrating its ability to learn on its own.

While cleaning a wash basin might seem like a simple chore, there’s actually a lot involved. The robot must intuitively understand the correct angle for the sponge, determine how much pressure to apply to different parts of the sink depending on the dirt, and continuously adjust its movements as it covers the entire surface. For humans, this is second nature, but for a robot programmer, it presents quite a challenge.

“Capturing the geometric shape of a washbasin with cameras is relatively simple,” says Andreas Kugi from the Automation and Control Institute at TU Wien in Austria. “But that’s not the difficult part. The real challenge is teaching the robot: Which type of movement is needed for which area of the sink? How fast should it move? What angle is best? How much pressure should it use?”

Realizing that programming all these variables would be a monumental task, Kugi and his team decided to take a different approach. Instead of coding every detail, they allowed the robot to learn by watching a human perform the task.

Robot Arm Trains to Clean Sink by Observing Human Actions with Specialized Sponge and Sensor Data

They created a special cleaning sponge equipped with force and position sensors, then had a person repeatedly clean the front edge of a sink sprayed with dyed gel to mimic dirt. The data collected from these cleaning sessions was used to train a neural network, which could then translate this information into specific movement patterns. The robot then applied these patterns to complete the cleaning task, and as shown in the video, it performed remarkably well.

While the experiment focused on sink cleaning, the researchers believe this method could be applied to a wide range of tasks, such as sanding, painting, or welding. Furthermore, through a technique called “federated learning,” robots could share what they learn with others. A fleet of robots could each gain local experience, then share their acquired skills to improve performance across all units.

“Imagine a network of robots working in workshops to sand or paint surfaces,” says Kugi. “Each robot could learn locally, and then share the parameters it learned with others.“

This advancement brings us a step closer to what many consider the future of robotics—and perhaps even a touch closer to the singularity.

A paper detailing this work was submitted to the IROS 2024 conference, where it won the “Best Application Paper Award” out of more than 3,500 submissions.

Read Original Article: New Atlas

Read More: Scitke