Research On Lateral Decision-Making Method For Intelligent Vehicles Imitating Driver Behavior

As an important part of future transportation,one of the key technologies of intelligent vehicles is the lateral decision-making system.This system integrates multiple core functions such as lane keeping,lane changing,and obstacle avoidance,and realizes safe and reasonable decision-making for vehicles through efficient and accurate environmental perception.The research on lateral decision-making systems is not only directly related to driving safety,but also the key to determining vehicle driving efficiency and passenger comfort,which is of great significance and application value.Although researchers have conducted extensive research on the key aspects of lateral decision-making,there are still many problems that need to be solved.Existing surrounding vehicle trajectory prediction models are often struggle to balance long-term effectiveness,real-time performance and accuracy,and perform poorly in dealing with complex and ever-changing traffic environments and driving behaviors.At the same time,the environmental risk assessment before decision-making needs to comprehensively consider the coupled interaction between people,vehicles and roads to achieve a comprehensive understanding and dynamic cognition of traffic elements.However,Traditional quasi-static risk models have problems such as weak prediction ability and over-idealization due to their over-reliance on instantaneous states.In addition,existing trajectory planning algorithms have made remarkable progress in pursuing safety,comfort and efficiency,but still need to be strengthened in the integration of driver behavior characteristics.The ultimate goal of the research is to ensure that the planned trajectories can closely meet the individual needs of drivers.In response to the above problems,this paper starts from the three aspects of lateral decision-making,namely surrounding vehicle trajectory prediction,environmental situation assessment and driving trajectory planning,and integrates the analysis of driver behavior characteristics into them,aiming to develop a more personalized lateral decision-making system for intelligent vehicles.The specific research contents are as follows:(1)Aiming at the problem of insufficient adaptability between lateral decision-making and driver operation of intelligent vehicles in complex traffic environments,the lateral behavioural characteristics of drivers are deeply analyzed and identified.A real vehicle data collection platform is built,and multiple drivers are recruited for natural driving data collection experiments.According to the process of time synchronization,data pre-processing and condition recognition,the key lane keeping and lane changing conditions are extracted from the original data,and based on this,the driver manipulation characteristics under typical conditions are analyzed.In order to describe the driver’s style representation,the factor analysis method is used to reduce the dimension of multi-dimensional feature variables,and then the k-means++ algorithm is used to achieve style clustering.In order to dynamically identify driver style in actual driving,an online identification method based on LSTM network is proposed,and the effectiveness of the algorithm is verified in the real vehicle scenarios.(2)Aiming at the problem of long-term and accurate vehicle intention recognition and trajectory prediction,a structured Informer model with intention perception ability is proposed.This model is based on an encoder-decoder framework,where the encoder is dedicated to identifying the intention of the vehicle,while the decoder focuses on predicting the trajectory.In the encoder stage,the model converts the location information of the target vehicle and interaction data with other vehicles into embedded vectors,and incorporates time-sensitive positional encoding to ensure that the timing information of the trajectory is completely preserved.The multi-layer self-attention mechanism is used to capture the social dependence and time dependence between successive frames,and finally output the intention probability distribution.The decoder part takes the masked trajectory and position encoding as input,and also utilizes the self-attention layer to handle the complex dependencies between features.By interacting with the encoder output,the decoder eventually generates the vehicle’s driving trajectory for the next 5s.By comparing with the current popular network models,the significant advantages of the proposed algorithm in prediction accuracy and computational efficiency are verified.(3)Aiming at the problem of dynamic quantitative expression of environmental risks,an interactive predictive risk field model based on dynamic environmental characteristics is constructed.After in-depth analysis of the interaction mechanism and risk influencing factors between people,vehicles and roads during vehicle driving,the risk field function modeling is carried out from three aspects: road conditions,driver characteristics,obstacle attributes and movements.In the process of constructing dynamic prediction field,the predicted trajectory of obstacle vehicle is introduced,and the shape and range of the field are adjusted dynamically based on the relative trajectory method,that improves the prediction ability of the field.Through the simulation experiment in the specific scenario and the comparison with the existing driving risk field models,the intuitiveness and comprehensiveness of the model in the quantitative expression of risk are demonstrated.(4)Aiming at the problem of low degree of personalization in current trajectory planning algorithms,a multi-objective trajectory planning method considering the driver characteristics is proposed.In Frenet coordinate system,the lateral and longitudinal motions of the vehicle are decoupled,and multiple candidate trajectory sets are generated based on the sampling and polynomial fitting of the initial and final states of the vehicle.Through trajectory inspection,some unreasonable trajectories are eliminated,which leads to the optimization of computational efficiency.A multi-objective evaluation system for trajectory clusters is designed from the dimensions of safety,comfort and efficiency.The fuzzy multi-attribute decision theory is used to analyze the subjective and objective evaluation of the driver,and calculate personalized index weights to select the optimal trajectory according to each driver’s preference.Simulation tests in multiple scenarios prove that the generated trajectories satisfy the basic objectives of safety,comfort and efficiency,and also meet the individual needs of drivers.(5)Aiming at the verification of the designed algorithm,joint simulation and real vehicle platform verification.A joint simulation platform is built based on Car Sim/Simulink/Prescan,and four typical scenarios are designed for simulation experiments.The hardware layout and software architecture of the two experimental vehicles are designed,and the real vehicle verification is carried out in three typical working conditions.The results show that the lateral decision-making system proposed in this paper can accurately predict the trajectory of the obstacle vehicle in various static and dynamic environments,reasonably quantify the potential risks in the environment,and plan a personalized driving trajectory that meets the expectations according to the driver’s preference for safety,efficiency and comfort.