Robots are going to become a whole lot more agile in the future. Plenty of researchers are working on perfecting them. Learning agile skills can be challenging but imitation learning can make it more easier. A team of researchers have introduced a Wasserstein Adversarial Skill Imitation (WASABI) approach that enables robots to acquire agile behaviors from “partial and potentially physically incompatible demonstrations.”
As this video shows, Solo, a quadruped robot, learned backflips from human demonstrations. Chenhao Li, Marin Vlastelica, Sebastian Blaes, Jonas Frey, Felix Grimminger, and Georg Martius are behind this approach.
[HT]