Imagine a world where robots clean your home, beaches, and even public spaces. We’re already well on our way there, but a new self-learning robotic arm developed by researchers at TU Wien in Austria has taken us a step closer. This robot has learned to clean a sink, a task that seems simple for humans but is quite complex for machines.
Cleaning involves more than just scrubbing. Humans instinctively know how much pressure to apply, which angle to use, and how to adjust as they go. For a robot, programming each of these movements would be a massive task. Capturing the shape of the sink is easy, the researchers say. However, teaching a robot how much force to use and which angle to clean at are entirely different matters.
To overcome this challenge, the team used a new approach. Instead of manually coding each action, they let the self-learning robotic arm learn by watching.
First, they designed a sponge with sensors that measure force and position. Then, a human cleaned a sink edge repeatedly, letting the sensors gather data. This data was used to train the robot’s neural network, allowing it to mimic human cleaning actions.
While this particular test focused on cleaning sinks, the researchers believe it could apply to more areas. The self-learning robotic arm could potentially handle various tasks on different surfaces, like sanding, painting, or welding. It could even possibly be attached to robot butlers that handle a variety of tasks in your home.
What’s more, this technology could enable multiple robots to share what they learn with each other through a method we call “federated learning.” Each robot could gain its own experience and then share what it learns, improving efficiency. Of course, that also sounds like the start of every robot uprising we’ve ever seen in the movies.
