Research On Trajectory Planning Algorithm Of Driverless Vehicle Considering Uncertainty

In recent years,unmanned vehicles have become a hot research direction to solve the important social problems such as traffic accidents,road congestion and energy crisis.There are uncertainties in the traffic environment of unmanned vehicles,which are objective,extensive and real,and only uncertainty is certain.Uncertainty will affect the risk assessment,intelligent decision and trajectory planning of unmanned vehicles,which will bring great challenges to the safety,reliability and comfort of unmanned vehicles.Therefore,this paper proposes a risk potential energy cloud model for uncertainty,which can well quantify the risk degree and probability of potential collision.Based on the risk potential energy cloud model,two different driving styles are constructed: high risk sensitive and probability sensitive,which provides more meta and personified intelligent decision-making for unmanned vehicles under different driving risk conditions and trajectory planning method.The main contents of this paper are as follows:(1)External environment perception for unmanned vehiclesIn this paper,YOLOv4 network is used to identify traffic participants such as vehicles,pedestrians and trees,lane line is detected by Lane Net network,and data association is used to fuse millimeter wave radar and camera data,so as to obtain the location,speed,attributes and other information of obstacles.(2)Build risk potential energy cloud model and intelligent decision system based on that modelConsidering the uncertainty of environment perception and kinematics in driving process,this paper superimposes normal distribution on the traditional risk potential energy field model to obtain the risk potential energy cloud model.In addition,different driving styles pay different attention to driving safety and traffic efficiency.Therefore,this paper divides driving style into high risk sensitive type and probability sensitive type,and builds two intelligent decision-making systems based on the risk potential energy cloud model.(3)Local trajectory planningFirst,the local path planning is decomposed into one-dimensional problems which are independent in horizontal and vertical directions by adopting Frenet framework,and candidate trajectory clusters are generated by quintic polynomial curves.In this paper,a trajectory evaluation function considering safety,efficiency and comfort is designed,and the candidate trajectory is evaluated to get the optimal path.(4)Simulation experiment verificationThis paper builds a Simulink/Pre Scan joint simulation platform,and designs five working conditions with different characteristics,and completes the simulation experiments of risk assessment,intelligent decision-making and path planning considering the uncertainty of the driving environment.The experimental results verify the safety,effectiveness and reliability of the algorithm in this paper.To sum up this paper obtains information such as the location and speed of obstacles through environmental perception,establishes a risk cloud model that considers uncertainty,conducts risk assessment and behavior decision-making based on the cloud model,and performs trajectory planning to obtain the optimal trajectory according to the decision results.