From cleaning urinals to cleaning up beaches, we can already see a future where robotic servants help keep the world a little cleaner. Now, a robotic arm has shown off its learning abilities by mastering a surprisingly complex sink-cleaning task.
Cleaning a sink may not sound like the most difficult task, but if you think about it, it’s a lot of work. You have to intuitively know at what angle to use the sponge, understand how much force to apply to different parts of the sink at different times, and constantly readjust your body as you move along the surface. It’s certainly easy for us humans, but if you’re a programmer working with a robot just starting out, there’s a lot of coding to do.
“Capturing the geometry of the sink with a camera is relatively simple,” says Andreas Kugi from the Institute for Automation and Control at TU Wien in Austria. “But this is not the critical step. It is much more difficult to teach a robot: what type of movement is needed on which part of the surface, how fast should the motion be, what is the right angle? What is the right amount of force?”
Understanding that programming all these data points and combinations would be a Herculean task, Kugi and his team decided to let the robotic arm learn how to perform the task by observing others doing it.
So they developed a special cleaning sponge equipped with force and position sensors, which they used to repeatedly clean just the front edge of the sink, sprayed with a dyed gel that mimics dirt. They then used the data collected from these movements to train a neural network that could translate the input into predetermined movement patterns. They provided these patterns to the robot and allowed it to signal its movements when it began its task. As you can see in the following video, the training went quite well.
Robot learns how to clean sinks
federated learning
The researchers say that although the experiment focused on cleaning sinks, it demonstrated that the robotic arm can perform a variety of tasks on a variety of surfaces, including sanding, painting, and sheet metal welding. Moreover, they say that a swarm of robots can learn basic movements from each other through ‘associative learning’ and then apply these movements to individual, designated tasks.
“Imagine many workshops using these self-learning robots to polish or paint surfaces,” says Kugi. “We can then have the robots gain experience individually from local data. Still, all robots can share the parameters they have learned with each other.”
Can you say “singularity”?
Documents describing the team’s work are available from TU Wien. It was recently submitted to the IROS 2024 conference and won the “Best Application Paper Award,” setting it apart from more than 3,500 other papers.
Source: TU Vienna
