The picture is familiar from baby horses or giraffes: when they are born, they stumble and stumble until they can stand and walk. A research team is now simulating this process with a robotic dog.
He’s not made of flesh and blood, but of cables, plastic and copper, but when he first tries to walk he makes young giraffes or foals look pretty old: The four-legged robot dog Morti doesn’t have a real head, is the size of a Labrador and is learning up and running in just an hour. A research team from the Max Planck Institute for Intelligent Systems in Stuttgart built it to explore how animals learn fluid movement sequences. They present their development in the journal “Nature Machine Intelligence”.
Although animals have all the necessary muscles and tendons for locomotion after birth, they first have to learn how to coordinate. The first attempts to walk therefore usually consist of uncontrolled walking around, the animals have to rely on their reflexes. In order to understand how the animals learn from stumblers, the Stuttgart researchers built the four-legged robot Morti. “As engineers or roboticists, we sought the answer by building a robot that has reflexes like an animal and learns from mistakes,” says former doctoral student at the Max Planck Institute, Felix Ruppert.
Result: Using an algorithm, the robot Morti learns how to use its mechanics as well as possible in just one hour. If he stumbles, the learning algorithm changes how far the legs swing back and forth, how fast they swing, how long a leg stays on the ground. Leg movement is adjusted until walking works without stumbling. “Changing what the legs are supposed to do is the learning process,” says Ruppert.
This is basic research at the interface between robotics and biology. In Morti, a computer does what nerve cells in the spinal cord do in animals. To do this, the robotic dog constantly compares sensor information from its feet with the movement patterns in its virtual spinal cord. “We cannot study the spinal cord of a living animal,” says co-author Alexander Badri-Spröwitz. “But we can model it in the robot.”
(This article was first published on Monday, July 18, 2022.)
