Study on the Minimum Diversion Decision Distance of Autonomous Vehicle on Freeway Based on Lane-Changing Trajectory Modelling

Most autonomous vehicles employ a relatively conservative lane-changing strategy in freeway system. In the diversion areas, autonomous vehicles typically initiate lane-changing to curb lanes at lower speeds at a considerable distance from the diversion point, resulting in a decrease in the overall traffic efficiency within the diversion areas. However, lane-changing decision points excessively close to exit ramps can exacerbate the urgency of the lane-changing process, prompting irrationally forced lane-changing and increasing the collision risk. To provide decision-making references for the safe and rapid diverging of autonomous vehicles in freeway diversion areas, this study proposes a minimum diversion decision distance (MDDD) model for autonomous vehicles through microscopic lane-changing trajectory data. Specifically, the lane-changing process was divided into waiting for the acceptable gap stage and executing the lane-changing stage in this model. Subsequently, UAV aerial photography and YOLOv5 target detection algorithm were used to measure traffic flow data and lane-changing trajectory data in diversion areas. The shifted 2nd Erlang model was used to fit the probability distribution of the time headway in diversion areas, and the acceptable gap waiting acceptable gap distance prediction model was constructed by analyzing the microscopic lane-changing behavior. A modified hyperbolic tangent model was utilized to fit the lane-changing trajectories of vehicles, used to estimate the distance for autonomous vehicles executing a lane-changing under different conditions. By defining the critical parameters of the model, the minimum diversion decision distance for autonomous vehicles for freeway with different design speed was obtained. The results indicate that the minimum diversion decision distance is related to the vehicle lane-changing trajectory and time headway distribution, and when the distance between the autonomous vehicle and the starting point of the ramp gradient section is farther than the MDDD, the diversion decision needs to be completed as soon as possible.