Self-learning robot mimics humans, cleans washbasins, completes multiple tasks
The robot was able to copy human “teachers” after being shown how to the do the task only once.
Thanks to researchers at TU Wein in Vienna, the promise of housecleaning robots is one step closer. The team has developed a self-learning robot to mimic humans to complete simple tasks like cleaning washbasins.
While this might sound mundane, the development is very significant as hard coding a robot to move a sponge over the complex curved edges of a washbasin would be a monumental task. To this end, the research team found a hack by blending observation with tactile data from human teachers to train robots to copy the same task.
This development will not only have applications in the home, either. The same learning process can readily be applied to many other tasks in the industry, like polishing surfaces, painting, sanding, or applying adhesives.
“Capturing the geometric shape of a washbasin with cameras is relatively simple,” says Professor Andreas Kugi from the Automation and Control Institute at TU Wien. “But that’s not the crucial step. It is much more difficult to teach the robot which type of movement is required for which part of the surface. How fast should the motion be? What’s the appropriate angle? What’s the right amount of force?” he added.
This is the same way human beings learn new tasks, especially skills like trades.
Learning by observing robots
“In a workshop, someone might look over the apprentice’s shoulder and say, You need to press a little 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 to let the robot learn [similarly],” he added.
As previously mentioned, the team did this by developing a special cleaning tool to help teach the robot; a sensor-impregnated sponge. Using this, human “teachers” used the sponge with force sensors and tracking markers to repeatedly clean the front edge of a sink.
“We generate a large amount of data from just a few demonstrations, which is then processed to help the robot understand what proper cleaning entails,” explains Christian Hartl-Nesic.
Using this data, the team used an innovative data processing strategy developed by the research team at TU Wien to enable the learning process.
Initially, the measurement data is statistically processed, and the results are used to train a neural network to recognize predefined movement elements, known as “motion primitives.”
This advanced learning algorithm then allows the robot to effectively clean the entire sink or other objects with complex surfaces after training, even after it has only been shown how to clean a single edge of the sink!
Article