Boston Dynamics' Spot is learning to do backflips, which is no easy trick. In this video, roboticist engineer Arun Kumar, part of Spot's behavior team, breaks down how he's researching and applying reinforcement learning to help Spot develop new behaviors, like an athletic stunt such as a triple backflip. He goes into why athletic stunts like flips matter for real-world performance, and how pushing Spot's hardware to the edge makes it a better robot for customers.