強化学習を用いた複数の目標値を考慮可能な車間距離制御システムの提案

抄録

The ACC (Adaptive Cruise Control) have been applied for many passenger vehicles generally. Even the conventional ACC has an enough performance to follow a preceding vehicle. On the other hand, it has a possibility to cause excessive acceleration and jerk. To avoid that excessive behavior, in this research, the RL (Reinforcement Leaning) which can consider some objectives to determine the control input is proposed for ACC. In addition, to balance performances of following to a preceding vehicle and reducing jerk, the RL’s reward is designed by using unique thresholds to this research. This method is evaluated by being compared with ACC by LQR (Linear Quadratic Regulator). As a conclusion, the balance of performances and the robustness to a dead time are evaluated for RL method.