Researchers Create Faster, Cheaper Way to Teach a Humanoid Robot to Walk Over Real-World Terrain

Across sand, soggy grass, and gravel. Up slopes and stairs and across level ground. Whatever uneven terrain Georgia Tech researchers could find on campus or easily simulate presented no problem for their two-legged humanoid robot.

Machine learning Ph.D. student Feiyang Wu led development of a new kind of whole-body controller that allowed the humanoid robot to traverse all those varied surfaces. His method is computationally faster and cheaper than the leading approaches for training robotic controllers. And he was surprised how well it worked, even on surfaces that weren’t included in the training.