While robots can be designed to be faster or stronger than animals, it's rare to meet a robot that's as resilient as an animal in the wild after it's been injured. Animals adapt to survive and researchers from France and the US are developing robots that can keep working even after receiving major damage.
A new paper published in the journal Nature details and demonstrates a new trial-and-error algorithm that enables a robot to adapt after being damaged, finding a way to compensate for its losses and continue to work, all within two minutes and without the need perform a self-diagnosis procedure or rely on pre-determined contingency plans. In other words, the robot is improvising on the fly, in the field, using the algorithm to more closely mimic how an animal might be forced to react when injured in the wild in order to survive.
Sick of Ads?
Join more than 500 New Atlas Plus subscribers who read our newsletter and website without ads.
It's just US$19 a year.More Information
After the robot is damaged it continues to attempt to perform its duties while taking in feedback on its own performance, creating what the researchers call a behavior-performance map that is constantly updating as the robot attempts to adapt in order to remain operational.
Watch the video below for an example of how it works – after losing power in one of its six legs, the first featured robot discovers a new gait within less than a minute of trail-and-error experimentation that will allow it to continue moving on its assigned trajectory.
The slow motion video section in the above is particularly interesting, showing that the robot almost appears to be performing a sort of power jump with its back legs to compensate and stay on track. The effect is similar to limping behaviors seen in humans and animals, allowing them to continue on following an injury rather than just spinning in a circle as many "dumb robots" might after losing power to a limb.
"Once damaged, the robot becomes like a scientist," explains lead author Antoine Cully. "It has prior expectations about different behaviors that might work, and begins testing them... Each behavior it tries is like an experiment and, if one behavior doesn't work, the robot is smart enough to rule out that entire type of behavior and try a new type."
The authors say that the new trial-and-error learning algorithm used by the robots could eventually lead to robots that are more independent and effective for applications like putting out forest fires and assisting in rescue operations without constant human supervision.View gallery - 2 images