Visuo-motor learning for face-to-face pass between heterogeneous humanoids

摘要: Abstract Humanoid behavior generation is one of the most formidable issues due to its complicated structure with many degrees freedom. This paper proposes a controller for humanoid cope this issue. A given task decomposed into sequence modules first, each which consists set module primitives that have control parameters realize appropriate primitive motions. Then, these are learned by sensorimotor maps between visual information (flow) and motor commands. The accomplishes selecting module, in selected advance. face-to-face ball pass RoboCup context chosen as an example task. (To best our knowledge, first trial.) corresponding approaching ball, kicking opponent, trapping coming player. In order show validity, method applied two different humanoids, independently, they succeed realizing more than three rounds.