Research And Implementation On Integration Navigation System Using RBFNN For Automotive Application In Complex Environment

Autonomous driving is a future direction of automobile industry.With the breakthrough of critical technologies of artificial intelligence,visual recognition,GNSS high-precision position and navigation,and 5G-V2 X communication,autonomous vehicles come to be a tendency for logistics and mass transportation demands.GNSS and INS have been widely used in vehicle navigation and location.Their stability would significantly increase while integrating them into a united system.The integration of GNSS/INS has emerged as a hot research topic in navigation technology.Generally,GNSS and INS have integration means as loose integration,tight integration and deep integration.The loose and tight integration focus on GNSS-assisted-INS navigation.The deep integration,however,emphasizes on deploying INS to assistant signal processing of GNSS baseband.Among them,GNSS/INS loose and tight integration could greatly improve integrated system reliability,which are especially suitable for the high dynamic vehicle applications in condition of the complex urban traffic.The deep integration system of GNSS/INS needs to adjust the software and hardware structure from the very bottom of GNSS receiver.In addition,the deep integration needs a further study in terms of its integration algorithm.In recent researches,relevant study on several aspects are very limited,such as the error model of deep integration GNSS signal tracking,the performance evaluation and optimization of GNSS baseband in deep integration system,and the anti-spoofing performance and reliability of integration system.Such case has constrained the further progress and application of GNSS/INS deep integration technology.This research is aimed to automotive applications in the mass market,according to the requirement of GNSS satellite navigation for autonomous driving,we focus on INS-assisted GNSS technology in a complex dynamic environment,proposing a solution for improving integration system reliability,completing the real-time integration system algorithm design,designing and realizing an integrated module.Furthermore,a vehicle integration system using the integrated module has been set up,along with many field tests for performance evaluation.The main work of this paper and the innovations achieved as follows:(1)Considering GNSS signal vulnerability for vehicle application,we propose anti-spoofing algorithm during the GNSS signal-tracking phase.We do some research on the error of INS that influence GNSS tracking loop in deep integration system,then build an error propagation model.Against those complex dynamic environments for autonomous driving application,the merits of deep integration tracking loop have been analyzed.In the following,the thesis proposes some effective methods for improving the reliability of vehicle integration system,from two research aspects of anti-spoofing and error analysis and mitigation.At last,we build the simulation and testing environment,verify the effectiveness of those proposed anti-spoofing algorithms and error propagation.(2)A GNSS/INS deep integration system model for on-board environment is proposed.In this thesis,a real-time vehicle integration model is set up.An INS algorithm for vehicle integration application is presented,in which a method of attitude updating combining quaternion method and rotation vector method is proposed.In addition,we complete GNSS/INS algorithm and system design,and propose a low-delay filtering algorithm suitable for embedded system.The feasibility of all of these algorithms has been verified in real application scenarios.(3)We design a real-time vehicle GNSS/INS integrated filtering algorithm RBFNN(Radial Basis Function Neural Network).Aiming at high precision vehicle application under the condition of the weak GNSS signals,the thesis propose an integration navigation method based on RBF neural network assisted filtering.We design the tight integration navigation system by employing the RBF neural network aided filtering technique.When GNSS signal is intact,the RBF neural network is trained.While the GNSS signal is of outage,the RBF neural network predicts the measurement input of integrated filter update process based on former training result.Furthermore,a mathematical model and algorithm based on RBF neural network and the adaptive Kalman filter are proposed.The superiority of the proposed model is verified by the practical experiments.(4)Design and realize a GNSS/INS integrated navigation module.We fulfill a real-time GNSS/INS integrated navigation module in the thesis and integrate the integrated navigation module into a vehicle navigation terminal.It mainly consists of an integrated hardware platform design,a real-time software system based on the hardware platform,the engineering design and implementation of function modules,and the design of navigation terminal and its reliability testing for vehicle lane navigation.Moreover,a GNSS/INS integration on-board navigation system is built,and in various conditions and environment of automotive application,a serial of lane-level deep integrated navigation tests is carried on.