Collaborative Game Of Unprotected Steering For Intelligent Vehicles In Open Road Scenarios

Compared with restricted road scenarios such as standardized intersections and closed roads in parks,open road scenarios are characterized by complex driving environment,lack of signal control and loose driving rules.In such open road scenarios,the unprotected steering of intelligent vehicles has become an important safety concern for intelligent driving upgrading to advanced autonomous driving.However,the current mainstream solutions,such as pre-setting driving procedure rules,fail to fundamentally solve the problem.Unprotected steering is highly emergent and accidental in the open road scene.In addition,intelligent vehicles have limited perception of complex road environment,so it is difficult to make timely and effective prediction,leading to the public’s doubts on driving safety and other issues,which greatly affects the process of intelligent driving upgrading to high-level autonomous driving.With the continuous innovation of artificial intelligence technology,machine learning algorithms represented by BP neural network are widely used in the field of intelligent driving.It is worth noting that using intelligent vehicles to train and learn in various road scenarios provides a new way to solve the problem of unprotected steering.However,the existing solutions mainly focus on the specific road scene with relatively simple driving environment and clear driving rules,while the open road scene with relatively complex driving environment and relatively loose driving rules lacks in-depth research.To solve the above problems,this paper proposes a collaborative game method of unprotected steering for intelligent vehicles in open road scenarios.By depicting the driving behavior characteristics of different intelligent vehicles,the collaborative game behavior of unprotected steering is studied to improve the driving safety and traffic efficiency in open road scenarios.Specific research contents are as follows:(1)Classification and recognition of driving behavior characteristics.Combined with subjective and objective analysis methods,the characteristics of driving behavior data were analyzed and cross-verified,and BP neural network was constructed to classify and recognize the characteristics of driving behavior.Firstly,a simulated driving experiment is built,driving behavior characteristic data is collected and processed,and the driving behavior characteristics of experimental subjects are analyzed by clustering algorithm.Secondly,based on the multidimensional driving style scale,the subjective evaluation questionnaire is designed and the answer results are analyzed,and the subjective and objective analysis results are compared and verified.Finally,the BP neural network algorithm is used to classify the driving behavior characteristics,improve the efficiency of training by optimizing the learning rate reasonably,and prove the convergence of the improved training algorithm from a mathematical point of view.(2)Cooperative game without protection.Based on accurate identification and classification of driving behavior characteristics,a collaborative game method for unprotected steering in open road environment is proposed.Firstly,the internal logic relationship between the classification of driving behavior characteristics and the cooperative game of unprotected steering is expounded.Secondly,taking unsignalized intersection as the typical representative of open road scenario,the conflict decision conditions and the initial state of cooperative game are set up.Finally,the collaborative game model of the unsignalized intersection is constructed,and the driving income function is designed from the three aspects of safety,comfort and efficiency,so as to realize the collaborative game of unprotected turning,and the existence and uniqueness of the collaborative game model Nash equilibrium is proved from the mathematical point of view.(3)Virtual road simulation and semi-real vehicle sand table experiment.Combined with Prescan+Matlab simulation and sand table road experiment,the vehicle performance of unprotected steering in open road scene was verified.Firstly,combining Prescan and Matlab to build a virtual road scenario for unsignalized intersections,and analyzing the decision-making changes of intelligent vehicles with different driving behavior characteristics during unprotected steering.Secondly,the traffic performance of collaborative game in multi-vehicle road environment is analyzed by setting three indexes: average traffic time delay,total traffic time and conflict rate.Finally,the sand table road experiment was conducted based on the Jetson Nano intelligent car,and the operation results under the acceptable gap,conflict table cooperation,reinforcement learning and collaborative game models were analyzed to verify the performance of unprotected steering under the open road scenario.