Sure, evolution invented mammals hovering 200 meters through the air huge flaps of skin and 3 foot wide crabs climbing trees, but has it ever invented a four-legged animal with telescopic limbs? No, it does not. Biology can’t really work that way. But robots can certainly.
Meet Dynamic Robot for Embodied Testing, also known as ANIMAL, a machine there changes the length of the legs on the run– not to creep people out, but to help robots of all stripes not overturn so much. Writing Today in the journal Nature Machine Intelligence, researchers in Norway and Australia describe how they made the ANIMAL learn to lengthen or shorten the limbs to tackle different kinds of terrain. So when they unleashed the transformative robot in the real world, it used that training to effectively step on surfaces it had never seen before. (Ie failed to collapse in a pile.)
“We can actually take the robot, bring it out, and it will just start to adapt,” says computer scientist Tønnes Nygaard from the University of Oslo and the Norwegian Defense Research Institute, lead author on paper. “We saw that it was able to use the knowledge it had previously learned.”
Migratory animals do not have extendable limbs because, first of all, it is just not biologically possible. But it is not necessary either. Thanks to millions of years of evolution refining our bodies, humans, cheetahs and wolves move with incredible agility, constantly scanning the earth in front of us for obstacles as we run.
Robots, on the other hand, need help. Even a super-sophisticated machine like Boston Dynamics robot dog Spot have difficulty navigating complex terrain. Giving robots telescopic legs both improves their stability as they move across different surfaces and increases their energy efficiency. Stumbling around consumes a lot of battery power, and a fluttering robot can injure itself or people nearby. “I think it’s a particularly good idea to have a body that can be adjusted,” says Francisco Valero-Cuevas, an engineer at the University of Southern California. developing four-legged robots but was not involved in this new research. “It simply came to our notice then. An adjustable body creates a more versatile robot. ”
Nygaard and his colleagues trained DyRET by first literally building the experimental sandbox. In the laboratory, the long boxes were filled with concrete, gravel, and sand, representing a variety of terrain that the bottom might find in the real world. Concrete is the easy one – neat and flat and predictable. Stepping in sand is much more uncertain, as with each step the robot’s legs will sink in unique ways. Gravel is a physically hard surface, just like concrete, but it is also unpredictable, as the rocks can be displaced, which complicates the animal’s footfall. “By having the three terrain examples with different hardness and roughness, you get a pretty good reproduction of a kind of general interaction between the morphology or the body and the environment,” says Nygaard.