Proposal and evaluation of reference-free model predictive control incorporating human-like driving features

Autonomous driving is achieved through environment recognition, path planning, and path tracking control. To achieve safe driving, it is necessary to increase the performance of these functions; however, this can lead to increased computational complexity. These functions share computational resources on the vehicle and must be realized in real time to guarantee safety. Therefore, reducing the computation time is a key issue in autonomous driving. We propose reference-free model predictive control (MPC) that incorporates features of human-like driving; this method can reduce the computation time. The proposed method does not follow a unique optimal path; it travels within a certain safety region. Thus, switching the control objective from trajectory tracking to constraint satisfaction reduces computational complexity. Besides, an event-triggered MPC (ETMPC) framework has been adopted to further reduce the computation time. We implemented the proposed method and compared the results with those of the ETMPC. The results confirm that computation time can be minimized to 1/10 further compared with that of the other methods. ETMPC is sensitive to observation errors and has the drawback of increased computation time. To solve this aspect, the proposed method incorporated the reference-free approach. The performance of the proposed method was evaluated using the observation error because it was robust against observation noise by human-like features. As a result, it was possible to reduce the computation time to 1/30.