Research On Driverless Vehicle Path Planning Method

With the rise of artificial intelligence technology,intelligent transportation research has become a hot spot.Through the real-time collection,transmission,processing and release of traffic information,intelligent transportation can induce travelers to choose travel time,travel mode and travel path more reasonably according to their travel needs,so as to balance the traffic flow of the road network,effectively alleviate the problem of urban traffic congestion,and improve the efficiency of the road network.Driverless vehicle is an important part of intelligent transportation system,and the maturity and popularization of the former technology is of great significance to the latter.Driverless vehicles have very broad application prospects in various fields of society.From the technical point of view,path planning is the key technology for the realization of driverless vehicles,so the research on path planning is of great significance.This paper takes the vehicle path planning method as the research object.On the basis of summarizing the research results of driverless vehicles,path planning algorithm and short-term traffic flow prediction methods at home and abroad.Firstly,according to the road conditions and traffic conditions suitable for driverless vehicles,11 road indexes are selected,and the screening index system of prohibited sections is formulated.Based on the collected road data,the BP neural network optimized by genetic algorithm is used to select 28 alternative driving sections from 40 roads to form the alternative driving network.The previous path planning was carried out under the existing urban road network environment,but the road conditions in the existing road network are uneven,and the traffic safety of driverless vehicles cannot be guaranteed.The road network after security screening can ensure the security of the path planned later.Secondly,in order to accurately calculate the travel time of the vehicle from the current location to the destination from the current moment,this paper analyzes the influencing factors of travel time from four aspects: human factors,vehicle factors,road factors and environmental factors.Then,based on the practicability and simplicity,the travel time estimation model is constructed,which mainly considers the traffic flow,traffic density and vehicle speed of the road section.The travel time is expressed as non-congested road travel time and intersection queue delay time,and is simplified into a form only related to the unknown quantity of the inflow.Then,the bidirectional LSTM neural network model is constructed to predict the future time inflow,and the future time inflow is substituted into the travel time estimation model to obtain the dynamic travel schedule.Finally,a dynamic road resistance function is constructed to describe the time that vehicles may need to go through a road section.The experimental results prove that the predicted value of the bidirectional LSTM neural network is close to the real value,the prediction performance is stable,and it has high adaptability to different traffic flow conditions.It can accurately predict the travel time of the vehicle on the road segment in the future period in real time.Thirdly,in the simplified urban traffic network model based on the existing urban road network,taking the minimum travel time as the objective function,the starting point constraint and the ending point constraint as the constraint conditions,the classical Dijkstra algorithm is improved by using the method of dynamically limiting the search area,and the single driverless vehicle path planning is carried out in the peak period and flat hump period respectively.The planning results prove that the dynamic path planning algorithm can avoid congested road sections during peak hours,which verifies the effectiveness and feasibility of the algorithm.Finally,some driverless lanes are set up in the existing urban road network,and the reservation and allocation mechanism of the right of way is implemented for the driverless lanes.First,the capacity restriction multi-path allocation method is used to allocate the flow of driverless vehicles to the driverless lanes first.Then,according to the flow allocation results,with the goal of minimizing the total travel time of all driverless vehicles in the system,the improved genetic algorithm is used to plan the path of driverless vehicles.The change of the optimal path and travel time of the driverless vehicles at the same origin-destination before and after the setting of the driverless lanes,during the peak period and the flat hump period after the setting of the driverless lanes,when the driverless lanes are reserved and when the driverless lanes are not reserved is compared.It is found that at the same starting and ending point,the travel time is saved after setting driverless lanes;there is a difference between the shortest path of the driverless vehicles reserved for driverless lanes during the peak period and the flat hump period,and the driverless vehicles reserved for driverless lanes during the peak period may only reserve part of the driverless lanes,thus the travel time of the shortest path in the peak period is longer than that in the flat hump period;the travel time of the driverless vehicles reserved for driverless lanes is less than that without reservation.The planning results prove the benefits of setting up dedicated driverless lanes for driverless vehicles and verify the effectiveness and feasibility of the algorithm.