The more knowledge and computing power an autonomous, self-learning robot has, the better it performs, right? Not in a new study that looked at how simple a self-learning robot can be.
How do you get multiple bots to do something meaningful together – in this case move them around – with rules that are as simple as possible? The researchers wondered. They made little robots, gave them a few lines of computer code, and put them on a track one after the other. They then checked to see if the robots were able to figure out how they could make the best progress as a whole, without there being any communication between the robots or the centrally controlled robots.
Despite their limited computing power, robots have also proven to be very effective in responding to unforeseen circumstances. A narrowing of the trail or a broken neighbor didn’t stop them. The researchers also want to test the robots in an open space. And they are looking to see if the same is possible with flexible robots of different shapes. A starfish, for example, where the arms then have to figure out how they can crawl across the ground as a whole.
Eventually, they even hope to be able to integrate the mechanism into the material of the robot, without the need for a computer.
In this audio you can hear Luuk van Laake from AMOLF. Learn more about the research here: Self-learning robots work like a train. You can find the article here: Continuous learning of emerging robotics behaviors.
“Food expert. Unapologetic bacon maven. Beer enthusiast. Pop cultureaholic. General travel scholar. Total internet buff.”