Research On Driving Collision Avoidance Of Autonomous Vehicles Under Game Scenarios

Autonomous driving is the core technology of intelligent vehicles,and it is also the focus of research and concern in academic and business circles.Safety is one of the important research topics of autonomous vehicles.Human drivers perceive the current near and far driving environment information,preview the future driving trajectory,and timely adjust the strategy according to the dynamic changes of the surrounding environment,and give specific driving actions.Human driving is a real-time rolling,optimization and adjustment process.From the perspective of control theory,autonomous driving is a layered closed-loop system of "environment-decision-planning-control".The decision layer interprets the perceived information and gives the driving decision instructions.At present,the main challenge and difficulty are to make vehicles have the ability to interact with the surrounding vehicles and other agents in real-time.Therefore,understanding the driving behavior of human drivers in the real traffic environment and making the autonomous vehicles have the same interactive decision-making logic has become the focus of research.To improve the brain-like intelligence of decision-making of autonomous vehicles,this paper focuses on the driving collision avoidance problem of autonomous vehicles at typical signalless intersections under game scenarios.The main contents of this paper are as follows:1.To better deal with the random and complex interaction process in complex scenes and improve the degree of personification of autonomous vehicle decision-making.This paper proposes a non-cooperative dynamic game algorithm for autonomous vehicles at signalless intersections to study the decision-making behavior with interactive processing ability between multiple vehicles.Parameterize the physical quantity of motion.The game theory paradigm is used to solve the problem that the autonomous vehicle interacts with surrounding vehicles and makes decisions to allocate the right-of-way.The driving requirements are modeled and refined into four game payoffs: safety of integrating psychological expectation,smoothness considering comfort and fuel-saving,fast passing for road traffic efficiency,surrounding space considering vehicle spacing.Based on meeting the traffic requirements,the weight coefficient is introduced to represent the diversification and personalization of driving style,and the constrained optimization objective function of multi-agent game interactive decision-making is established.2.The payoff indicators of different dimensions are normalized.The subgame perfect Nash equilibrium solution of multi-vehicle non-cooperative dynamic game interactive decision-making in each stage is solved by using the backward induction method.The discrete optimal action of each vehicle is processed by cubic spline interpolation,and the piecewise expression of cubic spline interpolation is derived.The mechanism of agent vehicle joining or quitting the game in real-time is proposed.It can judge whether any vehicle has participated in the interactive game,and make switchable game modes comprehensively combining with whether any vehicle has passed the conflict area.This reduces the complexity of the game system,simplifies the calculation of the decision problem,and also realizes the lightweight of the game system.At the same time,the driving style index is established to identify the driving style,and the sequential action of each stage game is replanned in real-time.3.Through the co-simulation of MATLAB/Simulink and Pre Scan,the multi-vehicle noncooperative dynamic game interactive decision-making algorithm based on switchable game modes is simulated.The effectiveness of the proposed method is verified,the potential conflict is effectively resolved,and the effect of anthropomorphic decision-making and personalized driving is realized.