Boston Dynamics’ Atlas humanoid robot was shown performing more parkour in a new video, complete with somersaults, handstands, jumps, turns, twists and more. The team implemented a new optimization algorithm to transform high-level descriptions of each maneuver into dynamically-feasible reference motions, which were created using new techniques that streamline the development process. Read more for the video and additional information.
Atlas tracks the motions using a model predictive controller that smoothly blends from one maneuver to the next, and by using this approach, the team developed the routine significantly faster than previous ones. It was so successful that they managed a performance success rate of around 80%. Boston Dynamics may have released a commercial version of their Spot robot dog today, but it may be a while, if ever, until we see Atlas in stores.
