Research On Autonomous Vehicle Behavior Decision Based On Overtaking Frequency

Autonomous driving technology integrates artificial intelligence technology into the automotive industry,which can overcome the defects of traditional artificial driving vehicles caused by human factors such as safety hazards and limited efficiency,improve the quality of life of human travel and solve the key problems of urban transportation,of which driving decision is the most core part of the autonomous driving technology level and is a hot spot and difficult research point in the field of autonomous vehicles.The prerequisite for driving behavior decision is to deeply perceive and reasonably evaluate the information obtained from the perception layer and make effective decisions.However,while artificial vehicles can make reasonable decisions based on their own driving experience and speed perception,autonomous vehicles cannot accurately judge their own state in the local traffic flow due to the lack of quantitative indicators of the relative state differences between neighboring vehicles and their own vehicles,so the accurate assessment of their own state is an important part of autonomous driving decision making research and the main basis for making driving decisions.In this paper,based on extensive reading of relevant literature at home and abroad and in-depth analysis,we conducted a research on self-driving vehicle behavior decision based on Overtaking Frequency,and the main research results of the paper are as follows:1.For the problem of limited perceptual information in the context of singlevehicle intelligence for self-driving vehicles,a parameter for quantifying the difference between neighboring vehicles and self-vehicles,i.e.,overtaking frequency,is proposed,the application of which can add a new auxiliary information to the existing perceptual parameters such as speed,acceleration,and relative distance for the autonomous driving decision system.First,in order to meet the research demand of this paper to quantify its own state in the local traffic flow as the entry point,the concept of Overtaking Frequency is proposed by analyzing the traditional artificial driving vehicles to change the driving state inducing factors and considering the influencing factors of autonomous vehicles,which is defined as "the difference between the number of vehicles overtaking in the adjacent lane of the ego vehicle and the number of vehicles being overtaken within a fixed time window length ",and Overtaking Frequency definition formula is optimized based on the concept of sliding window method.Secondly,the test platform for obtaining real vehicle data was built,and the software for detecting lateral overtaking and overtaken vehicles on the upper computer was developed by using the fusion of millimeter wave radar and GPS.Then,and through the data collected by the real vehicle test as the support for selecting relevant parameters,the simulator quantitative test is designed,and the correlation between parameters such as the speed difference between the neighboring vehicle and the ego vehicle,the ego vehicle speed and the length of the time window are analyzed in depth.Finally,the analysis results are used to fit the parameters as a function and obtain the calculation method of speed difference and time window length,which improves the construction of dynamic function of Overtaking Frequency.2.Aiming at the problem of the uncertain influence of the driving state of the neighboring vehicle of the autonomous vehicle on the ego vehicle,a hierarchical progressive assessment method of the driving state of the autonomous vehicle based on Overtaking Frequency is proposed to solve the problem that the autonomous driving field cannot accurately grasp its own state.Firstly,the parameters for assessing the operating state of the autonomous vehicle related to the ego vehicle and the evaluation parameters under the influence of the surrounding vehicles are selected from the perspective of horizontal and vertical integration for the establishment of the safetyvelocity hierarchical autonomous driving state assessment level.Secondly,the influence of the front vehicle and neighboring vehicles on the autonomous vehicle under various working conditions is analyzed in depth for different scenarios,and the safety level of the ego vehicle is evaluated based on TTC and THW parameters,and the speed suitability of the vehicle under different scenarios is graded based on Overtaking Frequency on the premise of safety as the benchmark.Support Vector Machine(SVM),Long Short-term Memory(LSTM)and Bidirectional Long Short-Term Memory(BiLSTM)models were developed to evaluate the driving status of autonomous vehicles based on the data sets constructed from real vehicle tests,the validity of the evaluation method was verified by comparing the three models with and without introducing Overtaking Frequency as an observed variable.3.A decision making method with safety as the baseline condition and efficiency as the incentive is proposed based on Overtaking Frequency for the fuzzy decisionmaking motive problem of autonomous vehicles.Firstly,considering the influence factors of surrounding vehicles on the driving state of the autonomous vehicle,the reward and punishment mechanism of the autonomous vehicle is constructed by designing the reward and punishment function in reinforcement learning with safety as the precondition and efficiency as the incentive according to the results of hierarchical comprehensive state assessment.Secondly,the deep reinforcement learning framework based on Twin Delayed Deep Deterministic Policy Gradient(TD3)algorithm is built with acceleration,deceleration,and steering as decision outputs,and the framework is used to train the agent.Finally,the autonomous vehicle is able to make timely and effective decisions through simulation tests,and the effectiveness of the decision model is verified by the results of smoothness of the vehicle trajectory,speed error size,acceleration stability,and front wheel corner stability of the output,and the proposed decision model is of great significance to improve the rationality of the self-driving vehicle decision.This paper addresses the problem of insufficient perception of traffic flow state by self-driving vehicles,innovatively introduces the concept of Overtaking Frequency,proposes a hierarchical and progressive method to evaluate the driving state of selfdriving vehicles,establishes a driving behavior decision method with safety as the benchmark condition and efficiency as the incentive,provides a new quantitative index for self-driving vehicles to perceive local traffic flow,and improves the driving behavior decision validity and rationality.