Research On Collaborative Sensing And Obstacle Avoidance Strategy For Platoon Based Autonomous Driving

With the development of 3GPP V2 X technology,intelligent connected vehicles are constantly entering people’s daily lives.By using vehicle communication,computing,and sensing resources,this technology can achieve autonomous driving control of vehicles based on exploring the traffic environment of the road network,thereby effectively improving vehicle driving safety and road network traffic efficiency.However,currently,most autonomous vehicles driving on the road are based on independent sensing of individual vehicles for driving control,and their detection angles and ranges are easily affected by the obstruction of surrounding vehicles and roadside facilities,forming a blind spot for detection and causing potential traffic accidents.3GPP V2 X communication has the transmission characteristics of low latency and high bandwidth,which can be used to share and fuse real-time sensing data with different detection angles,ranges,and accuracy generated by neighboring vehicle sensors through multi-vehicle collaboration,effectively compensating for the shortcomings of single-vehicle sensing and ensuring the safety of vehicle driving.On the other hand,vehicle platooning,due to its compact inter-vehicle spacing,can greatly improve road network capacity and traffic efficiency,and has received extensive attention in the transportation industry.Considering that vehicles belonging to the same platoon can maintain a stable communication topology during driving,facilitating the interaction and fusion of sensing information between vehicles,vehicle platooning is becoming a key technical solution for multi-vehicle cooperative sensing.Currently,some scholars have carried out preliminary research on cooperative sensing of vehicle platoons,but there are still some unresolved issues.Firstly,the design of the platoon’s resource coordination architecture is not perfect,and the communication and computing resources of the vehicles are difficult to meet the low-latency requirements for processing cooperative sensing tasks.Secondly,in traditional sensing fusion technology,the consideration of sensing resources within the platoon is not sufficient,resulting in the inability to effectively integrate the multiple sensing resources of vehicles within the platoon,making it difficult to achieve the sensing accuracy and coverage range required for collaborative autonomous driving.During road driving,vehicles need to accurately perceive the surrounding environment and predict the trajectories of surrounding vehicles.However,traditional prediction algorithms trained with historical data will decrease in performance as traffic conditions change.Therefore,efficient real-time update strategies are needed to address the accuracy issues of long-term predictions.Finally,traditional multi-vehicle cooperative obstacle avoidance path planning methods ignore communication delays,and in emergency situations,delays can pose serious safety hazards.Therefore,it is necessary to further design corresponding interaction architecture and efficient and safe cooperative obstacle avoidance path planning algorithms to shorten the time for vehicles to reach trajectory consensus.In response to the above challenges,this thesis focuses on the research of collaborative sensing and obstacle avoidance strategies for platoon-based autonomous driving,with a focus on designing a resource collaboration strategies for platoon,platoon-based vehicle sensing collaboration strategies,intelligent trajectory prediction,and collaborative driving algorithms empowered by digital twin systems.The main research content of this thesis is divided into four parts.1)A platoon-based collaborative architecture for autonomous vehicle resources? 2)Collaborative sensing strategy based on platoon? 3)AI-based multivehicles trajectories prediction mechanism? 4)A collaborative obstacle avoidance strategy for teams empowered by digital twins.The first section investigates the resource coordination architecture of platoon-based autonomous vehicles.The architecture is based on 5G New Radio Side Link(NR SL)technology,which organizes connected autonomous vehicles driving on the road into an autonomous driving platoon.In this thesis,an optimization model of business-resource scheduling strategy is established based on the factual state of the platoon and the state of sensing,communication,and computing resources of the networked autonomous vehicles.It solves the problem of insufficient sensing of self-driving vehicles without the assistance of roadside facilities,and is able to guarantee real-time communication.As the platoon size increases,the optimization problem becomes NP-hard.Then,the genetic algorithm is conceived to readily solve the optimization problem and effectively unload the realtime business.The algorithm can also reduce the communication overhead as well as the computing delay of processing business.Simulation results show that the proposed management strategy in this thesis is able to makes full use of the resources to improve the safety of the platoon and the capacity of business processing.The second section investigates the platoon-based collaborative sensing strategy.A single self-driving vehicle suffers from the problems such as insufficient sensing equipment and blind spots,which are caused by other traffic blocks.Based on the platoon-based collaborative sensing architecture proposed in the first part of this thesis,a multi-vehicle cooperative sensing system is designed,which can integrate heterogeneous multi-source sensing resources on autonomous vehicles in the platoon,thereby improve the sensing accuracy and breadth.Meanwhile,according to the requirements of the platoon sensing,an algorithm for scheduling the heterogeneous multi-source sensing resources is designed,which is able to reduce sensing redundancy and communication overhead by satisfying the sensing accuracy,thereby improving the driving safety of autonomous vehicles.The third section investigates the AI-based multi-vehicle trajectory prediction mechanism.One of the most important tasks for autonomous vehicle sensing is object tracking,i.e.trajectory prediction.After using the historical trajectory to train the prediction algorithm,the real-time sensing data cannot be used to update the prediction algorithm,which leads to the prediction performance degradation or even failure of the prediction algorithm.Therefore,this thesis proposes a trajectory prediction algorithm based on Li-GRU(Light weight Gate Recurrent Unit)neural network,which has the advantages of high precision and fast convergence.Meanwhile,in order to optimize the communication and computing overhead brought by updating the neural network,this thesis proposes a reinforcement learning based neural network update strategy,which reduces communication and computing overhead by ensuring the trajectory prediction accuracy.The fourth section investigates the collaborative obstacle avoidance strategy of the platoon with the DT system.When autonomous vehicles face unexpected road obstacles,they need to quickly carry out collaborative obstacle avoidance path planning.Existing researches neglect the impact of communication on obstacle avoidance performance.In this thesis,a platoon-based collaborative autonomous driving architecture for obstacle avoidance is proposed,which can perform cooperative autonomous driving on complex and curved roads without the RSU assistance.Meanwhile,this thesis designs a collaborative driving DT system,which deduces and optimizes the driving path according to the driving state of the vehicle and the surroundings.Finally,an emergency-based cooperative automatic driving algorithm is deployed in the system,which significantly improve the successful passing opportunity and reduce communication overhead.