‘Blind’ Robot Learns to Climb Stairs Independently

One of the main challenges facing roboticists is designing machines that can routinely handle things we tend to take for granted: learning new information, gripping strange objects, or navigating new landscapes — such as, for example, stairs. Today, watching a robot go up and down stairs is, well, relatively routine. Those robots, however, are relying on cameras and sensors to identify and handle potential obstacles. Because it’s a tricky problem, researchers from Oregon State University thought it’d be useful to develop navigation technology that doesn’t solely rely on vision systems. So, Engadget reports, they taught a bipedal robot how to climb stairs while, effectively, “blind.” Engineers developed a Reinforcement Learning technique and ran a robot through a number of potential situations in a simulator — ensuring they wouldn’t break the robot during the trial-and-error phase.