| As one of the most important ways to meet people's travel and transport needs,vehicle has always been an important field for the application of the latest technology.And as the high-tech clusters of new rise industrial revolution,the intelligence technology is naturally applied in the field of automobiles.Autonomous vehicle,in particular,comprehensively upgrades the traditional vehicles,reduces the manpower output of drivers to vehicles and provides a higher quality thorough service for drivers by integrating and using advanced electronic sensors,low delay network communication equipment,computers with high computing capability,electronic control actuators,etc.The generalization of the concept of autonomous vehicle is a synthesis composed of driving automation system,intelligent interaction system,entertainment system,security defending system and vehicle dispatching system.However,whether in terms of the complexity and value of the core technology or the most basic function of vehicle,driving automation system is the most important part of autonomous vehicle.Autonomous navigation is the key of the driving automation system.The completion status of the autonomous navigation in complex environments is the core evaluation standard of the intelligent level of autonomous vehicle.The task of this dissertation derives from National Natural Science Foundation Project(No:51875061),key support project of Automobile Industry Innovation and Development Joint Fund of China(No:U1564208)and key technical research project of Changan Automobile.It investigates the generic core technology of the autonomous navigation in complex environments.The research has focused on clarifying the understanding framework of complex environments,both theoretically and experimentally.Meanwhile,this dissertation also discusses the full-size association of multidimensional environmental information in complex environments and the reconstruction of environments.Furthermore,The research studies the sparse representation method for structural point clouds and its application in the autonomous navigation.It also explores the self-adaptive path planning for unstructured environments.The study aims to guide the product design,platform constructing and engineering application of autonomous navigation in multiple operating conditions and multi-scenarios.The main details of this study is as follows:The simultaneous simplified model of environment-vehicle-sampling coordinate system is established.Firstly,the study juggles environmental coordinate system,vehicle coordinate system and sampling coordinate system.Secondly,it deduces the simultaneous simplified model of environment-vehicle-sampling coordinate system by making use of the geometric transformation of rigid body in the space,steering geometry model of vehicle and differential equations of vehicle motion.Then,according to the layout of Lidar and the unified assumptions of the driving automation system,the simultaneous simplified model is reasonably simplified.These simplified models are the geometric basis of the autonomous navigation system based on structural point clouds.Subsequent studies are based on these simplified models.The study explores the environment reconstruction method based on the signal structure transformation results of multi-dimensional signal of structural point clouds.First of all,the joint calibration method of vehicle-GPS-Lidar is introduced.Secondly,with the sampled signal of point clouds,the structural point clouds are separated from the background.Meanwhile,the structural point clouds are transformed into a signal structure in which full-size environmental information can be associated in the same basis.Then,static single-frame of structural point clouds is ultimately obtained.Finally,according to the static single-frame of structural point clouds and the simplified models of environment-vehicle-sampling coordinate system,the environments are reconstructed.The multi-dimensional signal sparse representation method of the environmental structural point clouds is derived.Using the environmental structural data,multi-dimensional signals are down-sampled to get the redundant point clouds before the sparse representation.The different domains data of multi-dimensional signals are separated into the source signals of the sparse representation.The kernel function of the over-complete dictionary is obtained by deducing the transformation of the orthogonal basis.With the introduction of the frequency factor of the dictionary generation,the over-complete dictionary is set up.The signal decomposition and reconstruction methods,which are suitable for the sparse representation of multi-dimensional signals of environmental structural point clouds,are described.The local path planning in complex environments is discussed.The local path planning algorithms which are suitable for structured environment are described in detail.Two kinds of existing local path planning algorithms for unstructured environments are derived,one is based on path search mode,the other is the self-adaptive path planning algorithm for unstructured environments.On this basis,a self-adaptive path planning algorithm for unstructured environments,path planning with dynamic potential field,is proposed.The motion planning of driving automation systems in unstructured environments,integrated 3D point clouds sparse representation and potential field(3DPCSR-PF),is proposed.In this algorithm,the sparse representation of the environmental structural point clouds is integrated with the repulsive force calculation of dynamic potential field,which can reduce the transmission and storage cost.Real vehicle tests examine the theories of this study,and the results of the tests are analyzed.Above all,the real vehicle test hardware platform is built,according to the simplified strapdown model of vehicle-sensors and the actual scale of candidate test vehicle.The structural composition of the complex environments is analyzed in detail,the specific test environments that meet the requirements are selected.The signal structure transformation of multi-dimensional signal of structural point clouds and the environment reconstruction are verified by real vehicle tests,and the results of the tests are analyzed.The corner section with dense obstacles is selected as the test environments,the multi-dimensional signal sparse representation method of the environmental structural point clouds is verified by real vehicle tests,and the results of the tests are analyzed.The representative road sections with different environmental structures are selected as the test environments to verify the dynamic potential field algorithm and the 3DPCSR-PF.The 3DPCSR-PF and RRT* algorithm are used to plan the feasible path with the same starting and target in same test environments,then the performance of their planning is compared.The influence of moving obstacles and self-vehicle scale on 3DPCSR-PF algorithm is analyzed through real vehicle tests. |
PDF Full Text Request