It’s been a year and a half since we watched Jean-Baptiste Mouret’s hexapod teach it self how to walk again after losing a leg. Damage resilience is certainly critical to being able to remain mobile while exploring that scary and dangerous world outside of a laboratory, but adaptation can be taken even farther, and Mouret’s new robot is exploring exactly how far that is on six adaptable wheels.
On flat ground, the Creadapt robot can use its wheels. But as soon as it encounters some other surface, like dirt or grass or rocks or rubble or mysterious grey goo, it’ll change its gait to whatever is most efficient. How does it know what the most efficient gait is for a given surface? It doesn’t. Yet.
Fundamentally, a robot losing a leg and a robot trying to move across a foreign surface efficiently is the same problem: the robot has to adapt to a new environment. To do this, the Creadapt project is teaching robots to do what animals (like humans) do. Namely, we innovate, optimize, evolve, and adapt. Or in other words, we try different things until we find what works, and then we keep trying variations on what works until we’ve found what works best.
The overall purpose of [the Creadapt] project is to fulfill this gap [in evolutionary robotics] by employing both the creative and the adaptive abilities of evolutionary algorithms to design algorithms that can autonomously and creatively adapt the behavior of robots to unforeseen situations. In the typical scenario, a mobile robot faces a situation that requires adaptation (e.g. a leg is broken or the ground surface changed). The robot is allowed to launch a few experiments to investigate the situation; after a few minutes it should be able to cope with the new situation to pursue its mission until a new adaption is required.
The general philosophy here is to teach robots to be able to find their own solutions to complex problems. This is important for long-term reliable autonomy, because there’s no way that we can possibly foresee all of the different problems that a robot might run into. And if you have some robots that are, say, exploring another planet and, I don’t know, one of them gets stuck in a sand dune or something, you’d want it to be able to try stuff until it manages to get free, learning from each attempt as it does so.
Creadapt has funding through 2016, so we’re just starting to see what might be possible here. Eventually, algorithms that allow robots to adapt and recover could be everywhere, making them much more reliable than ever before.
And very, very hard to stop.
[ Creadapt ]