And Robots that Learn…
Another team in Paris also wants to give robots further autonomy by teaching them how to learn and adapt. Antoine Cully, Jean-Baptiste Mouret and Danesh Tarapore of UPMC (formerly Universite Pierre-et-Marie-Curie), in collaboration with Jeff Clune of the University of Wyoming, are a team of robotic and software engineers who were inspired by animals’ ability to adapt to injury. Their research cites three-legged dogs and humans with sprained ankles who learn to adjust their walk.
“When injured, animals do not start learning from scratch,” said senior author Jean-Baptiste Mouret. “Instead, they have intuitions about different ways to behave. These intuitions allow them to intelligently select a few, different behaviors to try out and, after these tests, they choose one that works in spite of the injury. We made robots that can do the same.”
Much like the team in Brussels, Mouret’s team focused on the fragility of robots, in this case their inability to recover from damage via compensating and improvising behaviors. Their solution was to introduce a trial-and-error algorithm into the software of two robots that gifts them with the ability to adapt.
The algorithm works in two steps. First, it builds a map of all possible behaviors, movements and performances in a designated space and sets this information aside as “prior knowledge.” In the event of an injury, the algorithm uses the prior knowledge to enable the robots to adapt and compensate. The team demonstrated the effectiveness of the algorithm by breaking two legs of a six-legged robot, and several motors on a robotic arm. Both were still able to complete their assigned tasks in a matter of minutes. Additional experiments of various types of broken, damaged and missing components showed successful adaptation.
“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,” said lead author Antoine Cully. “For example, if walking, mostly on its hind legs, does not work well, it will next try walking mostly on its front legs. What’s surprising is how quickly it can learn a new way to walk. It’s amazing to watch a robot go from crippled and flailing around to efficiently limping away in about two minutes.”
…Give Rise to the Autonomous Machine
Tarapore believes that continued study of injury adaptation and further development of the trial-and-error algorithm will reduce the continuous attention robots need for repair, making them more productive and effective, and less of an annoyance. The endgame, of course, is the rise of more autonomous robots.
Professor Bram Vanderborght, one of the researchers on the Brussels team, is also excited by the prospect of an industry of independent machines.
“The outcome of the research opens up promising perspectives. Robots can not only be made lighter and safer, they will also be able to work longer independently without requiring constant repairs.”