Research on Vehicle Mandatory Lane Change Decision Model Based on Trajectory Analysis

Fang Zhu,Hanqing Xia,Zechao Ma,Lin Zhang,Yingjie Zhao,Yuanrong Wang
DOI: https://doi.org/10.1117/12.3039907
2024-01-01
Abstract:A mandatory lane changing decision model for self-driving vehicles is proposed for forced lane changing constrained by objective conditions. First, the NGSIM dataset is analysed to extract the trajectories of completed mandatory lane changing and the trajectories of uncompleted mandatory lane changing as the training data of the lane changing decisionmaking model, and for the problem that it is difficult to obtain the trajectory data of the trajectories that have the intention of mandatory lane changing but have not succeeded in changing lanes, this paper puts forward a method of extracting the target trajectories through the transverse acceleration thresholds, and the lateral acceleration thresholds are obtained from the vehicle's mandatory lane changing through the DBSCAN clustering. The transverse acceleration threshold is obtained by DBSCAN clustering, and the unsuccessful vehicle trajectory is extracted from the trajectory of the completed forced lane change according to this threshold. Finally, based on the Transformer encoder unit, the decision-making model of forced lane changing for self-driving vehicles is constructed. The model is verified to be able to achieve accurate decision-making for forced lane changing, which illustrates the reasonableness of filtering the unfinished trajectories of forced lane changing based on the transverse acceleration threshold, and the accuracy of the forced lane changing model reaches 95.6%, which is better than that of the long and short-term memory LSTM network model and the gated recurrent unit GRU model.
What problem does this paper attempt to address?