Like a newborn animal, a four-legged robot stumbles around during its first walking attempts. But while a foal or a giraffe needs much longer to master walking, the robot learns to move forward fluently in just one hour. A computer program acts as the artificial presentation of the animal’s spinal cord, and learns to optimize the robot’s movement in a short time. The artificial neural network is not yet ideally adjusted at the beginning, but rapidly self-adjusts. On July 18, the research work which is at the intersection of robotics and biology will be published in the renowned journal Nature Machine Intelligence.
Robot dog learns to walk in one hour
Virtual spinal cord is continuously optimized