| There are still many challenges to achieving fully autonomous driving on urban roads,and L3 autonomous vehicles will be around for a long time.In particular operating conditions where the driver is required to monitor and take over the vehicle,the issue of takeover safety has attracted much attention from researchers.Takeover time is a key factor in measuring the safe driving of L3 autonomous vehicles.Based on the analysis of existing studies,the following issues are identified:The interaction mechanism between L3autonomous vehicles and the surrounding traffic on takeover time is not sufficiently studied;The spatiotemporal variation relationships of the variables are ignored in the takeover time prediction problem;And the trajectory planning considering the takeover time factor is not well studied.Therefore,this study utilizes Cat Boost and SHAP models to reveal the influence mechanism of self-vehicle driving state factors and surrounding traffic flow factors on takeover time.Based on the analysis of influencing factors,the takeover time of L3 autonomous vehicles is modeled and predicted.Further trajectory planning research is conducted based on the predicted takeover time,which is crucial to improve takeover safety and stability.The details of this study are as follows:(1)Analyzing key influencing factors of takeover time and revealing the influence mechanism of surrounding traffic on the takeover time of L3-AV.In this study,an explanatory and visualization analysis of the actual and hardware-in-the-loop takeover data is carried out separately using Cat Boost as the prediction model and the SHAP model to explore the mechanism of each variable on takeover time.The results show that the self-vehicle driving speed is the main influencing factor of takeover time in the actual takeover experiments.Drivers pay more attention to vehicles coming from the opposite direction in the front left during the takeover.The relative feature importance of the interaction factor between the minibus and the surrounding traffic is the greatest.In the hardware-in-the-loop takeover experiments,the heading angle of L3-AV has the greatest effect on takeover time,and the driver focuses more on the left rear and front vehicles during the takeover.The relative feature importance of the interaction factor between the L3-AV and the surrounding traffic is the greatest.Compared with the actual takeover experiments,the hardware-in-the-loop takeover experiments focus more on the driving states of the surrounding traffic.(2)Modeling and predictive analysis of the L3-AV’s takeover time based on deep learning algorithms.Based on the influencing factors analysis,the takeover time is modeled and predicted using the ETSformer model and compared with baseline algorithms to verify the model prediction performance.The results show that the ETSformer model can accurately predict the takeover time of L3-AV,with R~2values of 0.90,0.83,and 0.93 for the prediction results on the fused takeover dataset,the actual vehicle takeover dataset,and the hardware-in-the-loop takeover dataset,respectively.The algorithm outperforms other comparative algorithms for takeover time prediction,with a mean absolute percentage error of 7.2%for the fused takeover dataset,a 0.7%reduction relative to the best-performing baseline algorithm.The error is minimized when the self-vehicle driving state factor,the traffic flow state factor,and the surrounding vehicle characteristics are simultaneously considered in the takeover time prediction.In addition,the takeover time prediction results have an important impact on the trajectory planning of L3-AV,especially in complex scenarios where the takeover time factor must be considered for safe trajectory planning.(3)Constructing an L3-AV trajectory planning framework based on takeover time prediction and validating and evaluating the results.An L3-AV trajectory planning framework based on takeover time prediction is constructed,including surrounding vehicle trajectory prediction considering spatiotemporal interaction,lane change decision,and L3-AV trajectory planning based on lane change decision and takeover time prediction.Minimum time to collision,lateral and longitudinal acceleration and braking measures,and maximum lateral offset are applied as evaluation indicators.The trajectory planning effectiveness of the proposed framework is verified based on the OpenCDA framework.The results show that planning trajectories based on takeover time prediction can achieve safe obstacle avoidance after the takeover and exhibit better stability and reliability under different conditions compared to real trajectories and planning trajectories without considering takeover time.The takeover time has an important impact on the safety and stability of the driving trajectory,and a lengthy takeover time is more detrimental to the L3-AV’s trajectory planning. |
PDF Full Text Request