AbstractAdaptive cruise control (ACC), which is an extension of conventional cruise control, has been applied in many commercial vehicles. Traditional ACC is controlled by proportional-integral-derivative control or linear quadratic regulation (LQR), which can provide sufficient performance to follow a preceding vehicle. However, they can also cause excessive acceleration and jerk. To avoid these excessive behaviors, we propose reinforcement learning (RL), which can consider various objectives to determine control inputs, as an ACC controller. To balance the performance of following a preceding vehicle and reducing jerk, RL rewards are designed using unique thresholds. Additionally, to balance performance and robustness to the zero-order delay (dead time) of the controlled system, dead time is also considered by scattering it randomly in the learning phase. As a result of this study, an RL agent trained using the proposed RL method and two LQR units specialized for followability and comfortability were simulated using Simulink® (MATLAB®).