PERFORMANCE EVALUATION OF AUTONOMOUS VEHICLE CONTROL STRATEGIES FOR LEFT-TURN MANEUVERS AT SIGNALIZED INTERSECTIONS

<p dir="ltr">Autonomous vehicles (AVs) have revolutionized urban mobility by enhancing safety and efficiency. However, complex maneuvers such as left turns at signalized intersections pose significant challenges. These intersections are among the most hazardous zones, with left turning vehicles involved in about 21% of fatal intersection crashes, highlighting the urgent need for reliable control strategies. This study addressed this need by focusing exclusively on the control layer and conducted a simulation-based comparison of two decoupled control architectures: Proportional-Integral (PI) and Stanley control (PI + Stanley), against Model Predictive Control (MPC) and Stanley control (MPC + Stanley). This research also assumed a pre-generated, collision-free reference trajectory, with vehicle motion simulated by a bicycle model operating as a rigid body with ideal actuators. The primary objective was to assess how these architectures compared in terms of safety, comfort, and computational cost when tracking the reference trajectory during time-critical left turns. By comparing crucial performance metrics across multiple simulation runs, this study found that the MPC + Stanley Controller performed better in terms of safety and comfort whereas the PI + Stanley controller posed a lower computational cost. The performance metrics used to quantify the comparison were the cross-track error, heading error, and speed tracking error for safety; jerk and trajectory smoothness for comfort; control loop frequency; and computational cost. These findings contributed to safer AV deployment in dynamic signalized urban environments.</p>