2103 強化学習を用いたマルチロータヘリコプタの行動学習(OS10-1 ビークルオートメーション,OS10 アドバンティシンポジウム,オーガナイズド・セッション(OS))

Abstract

We tried to control quadrotor helicopter which is one of the multirotor helicopters by reinforcement learning. Quadrotor helicopter is a very simple machine which consists of four rotor blades and rigid cross airframe. However, the nonlinear dynamic behavior of quadrotor helicopter requires a more advanced stabilizing control. On the other hand, reinforcement learning agent acquires optimal action by itself for a purpose of maximizing reward through trial and error. In this paper, we try to apply Q-learning to quadrotor helicopter in order to easily control. We verified that quadrotor helicopter with Q-learning could acquire optimal action by itself to keep target altitude in flight experiment.