深層強化学習による四足ロボットの歩行動作獲得

抄録

The declining working population due to the aging of society has prompted the introduction of quadruped robots. However, it is difficult for quadruped robots to coordinate and control their motors. Therefore, machine learning, in which the robot learns movements autonomously, is attracting attention. In addition, quadrupeds have different movements when they walk and run. In this study, we investigated whether the robot acquires different movements depending on the target speed using deep reinforcement learning. As a result, the robot acquired walking motions such as "pace" and "gallop" as in animals.