Watch a shape-shifting robot prowl the big, bad world
Certainly, evolution was invented mammals that float 60 meters through the air giant flaps of skin and 1 meter wide crabs who climb trees, but did it ever invent a four-legged animal with telescopic limbs? No it has not. Biology can’t work like that. But robots certainly can.
Meet the Dynamic Robot for Embodied Testing, also known as DyRET, a machine that changes the length of his legs while flying– not to huddle humans, but to help robots of all shapes and sizes not fall over so much. To write today in the diary Nature Machine IntelligenceResearchers in Norway and Australia describe how they got DyRET to learn how to lengthen or shorten its limbs to tackle different types of terrain. Once they released the shape-shifting robot into the real world, he used that training to efficiently tread surfaces he’d never seen before. (That is, it failed to collapse into a heap.)
“We can actually take the robot, bring it out, and it will just adapt,” said computer scientist Tønnes Nygaard of the University of Oslo and the Norwegian Defense Research Institute, the paper’s lead author. “We saw that it could use the knowledge it had previously acquired.”
Walking animals don’t have extendable limbs because it just isn’t biologically possible in the first place. But it is also not necessary. Thanks to millions of years of evolution that sharpen our bodies, humans, cheetahs and wolves all move with incredible agility, constantly scanning the ground in front of us for obstacles as we run.
Robots, on the other hand, need a little help. Even a super advanced machine like the Boston Dynamics robot dog Spot has difficulty navigating complex terrain. Giving robots telescopic legs improves their stability as they move across different surfaces and increases their energy efficiency. Tripping takes up a lot of battery power, and a swinging robot can injure itself or people nearby. “I think it’s a really good idea to have a body that’s adjustable,” said Francisco Valero-Cuevas, an engineer at the University of Southern California. who develops four-legged robots but was not involved in this new research. That’s what’s happening here. An adjustable body makes for a more versatile robot. “
Nygaard and his colleagues trained DyRET by first literally building experimental sandboxes. In the lab, they filled long boxes with concrete, gravel, and sand, representing a range of different terrains that the bot could find in the real world. Concrete is the easy one: nice and flat and predictable. Stepping into sand is much more uncertain, as the robot’s legs would sink in unique ways with each step. Gravel is a physically tough surface, like concrete, but it is also unpredictable, as the rocks can shift, making DyRET’s footsteps more difficult. “By having the three terrain examples, with different hardness and roughness, you get a pretty good representation of some sort of general interaction between the morphology, or body, and the environment,” says Nygaard.