モデル予測制御を用いた二輪車両ロボットの障害物回避

―BUG2実装による停止点の脱出とNN制御則近似―

抄録

In this study, we attempt to drive a two-wheeled vehicle from a specified starting point to a target point avoiding obstacles by using model predictive control. In particular, we focus on dealing with the case of a deadlock due to an obstacle and on reducing the computational complexity. In the case of a deadlock, we modify the objective function of the model predictive control to get out of the deadlock. In order to reduce the computational complexity, the control law is approximated by a neural network.