A group of French and American scientists publishing in the science journal Nature are reporting success in their attempt to develop software that allows damaged robots to quickly overcome many sustained injuries and continue with their tasks.
A major problem in robotics has long been the inability of the machines to deal with unexpected circumstances, and in the case of damage suffered in the performing of their duties, they will usually just ignore the problem and thus make the it all the worse. Now, though, with inspiration drawn from nature itself, the scientists — Antoine Cully, Jean-Baptiste Mouret, and Danesh Tarapore of the Marie Curie University in France, and Jeff Clune from the University of Wyoming — have created software that allows robots to assess their own capabilities, test new movement tactics through trial-and-error, and decide on a going-forward solution, all in just a matter of minutes.
Antoine Cully, the lead author of the Nature piece (entitled "Robots That Can Adapt Like Animals"), writes, "Once damaged, the robot becomes like a scientist.", going on to say "It has prior expectations about different behaviors that might work, and begins testing them. However, these predictions come from the simulated, undamaged robot. It has to find out which of them work, not only in reality, but given the damage."
Specifically, with the new software a robot is able to construct a map of their surroundings and their position therein. Then, in the event of an injury or malfunction, the "Intelligent Trial and Error" algorithm engages, through which the robot is able to perform tests on itself and work out a solution that will allow it to continue its tasks.
Cully states, "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. 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."
The scientists are confident that their software solution can be of value to a large array of different robot types, providing means by which the machines can achieve greater margins of success, continuing to perform their work until such time as repair is reasonable and convenient.
"It could enable the creation of robots that can help rescuers without requiring their continuous attention," says co-author Danesh Tarapore. "It also makes easier the creation of personal robotic assistants that can continue to be helpful even when a part is broken."