Research On Self-Contained Integrated Navigation Algorithm Based On Low-Cost Sensors For Autonomous Vehicles

An accurate and reliable navigation system is essential for an autonomous vehicle.The global navigation satellite system based on radio is prone to be jammed because of shelters in urban canyons.In this case,its precision will degrade for a short time.Perhaps worse,there will be an interruption in navigation information.To solve this problem,self-contained integrated navigation technologies based on low-cost sensors are researched in this paper.Combining SINS that can provide short-term high-precision navigation information with other measurements those can provide space constraints,two integrations are studied to achieve a continuous and more accurate navigation solution.Specifically,one is the VKM/SINS integration,and the other is the monocular-vision/SINS integration.The main contributions include the following aspects.1.Basic theories of self-contained integrated navigation technologies based on low-cost sensors are analyzed including SINS and its error propagation model,multiple view geometry in computer vision,Ackerman steering principle and Kalman filter.2.A VKM/SINS integrated navigation algorithm is proposed.Compared with traditional method,this algorithm that utilizes multiple homogeneous kinematic constraints as velocity measurements is more accurate and robust.Furthermore,in order to enhance the performance,an adaptive fusion method with fault detection is adopted.Matlab/Simulink-Car Sim co-simulation is used to research the performance of this algorithm.As a consequence,low positioning error drift of the proposed algorithm is verified,and the influence of horizontal misalignment angles is analyzed.3.Monocular-vision /SINS integrated navigation algorithms based on loosely coupled and tightly coupled architecture are studied respectively.In addition,benefiting from a scale estimation method for vehicles,a monocular VO that can provide metric positioning information is proposed.4.Experiments in real vehicular environments are used to test and verify the performance of the algorithms.A vehicle experimental platform in our group is utilized to evaluate the VKM/SINS integrated algorithm.Experimental results indicate that the RMSE of horizontal error ratio of the proposed method is less than 4‰,and the RMSE of height error ratio is less than 3.7‰.The KITTI dataset is used for the monocular-vision/SINS integration.Experimental results indicate that the RMSE of horizontal error ratio of the monocular VO based on scale-estimation is about 2.2%.However,due to the poor performance in velocity estimation,the loosely coupled method that utilize it as the observation cannot provide navigation information as expected.Tightly coupled method that combine the advantages of all sensors shows a steady navigation performance whose RMSE of horizontal error ratio is about 1.5%.