Research On Multi-Sensor Navigation And Lateral Robust Control Of Autonomous Vehicle

In order to satisfy the developing requirements of automobile road tests, several key technologies of a robot driver which is applied in automobile road tests are investigated in this dissertation. From the viewpoint of system engineering, the robot driver with the abilities of navigation, information fusion and control etc. and the automobile are regarded as an organic union, which is called an autonomous vehicle in this dissertation. To realize automatic driving of the autonomous vehicle on the structural test road, several navigation technologies, multi-sensor fusion and integrated navigation method, and lateral control etc. are studied. The main research work and fruits are summarized as follows:(1) According to the characteristics of inductive symbol lines on the structural test road, a real-time image detection method is proposed. To obtain the feature parameters of inductive symbol lines, three different methods are used separately and evaluated by experiments. Experiments in typical road environments demonstrate that the proposed detection method based on Kalman filter by scalar measurement processing exhibits good performances in accuracy, reliability and processing velocity.(2) A vision measurement method of lateral deviation error under the condition of near view is presented according to camera's perspective projection relationship. As to the camera calibration in this measurement method, a simple and feasible calibration method is adopted which can adapt to outdoor road environments. Experimental results demonstrate that the proposed vision measurement method of the lateral deviation error can achieve the cm-level accuracy on the general road under the condition of near view.(3) SINS/GPS integrated navigation of the autonomous vehicle is deeply studied. The SINS/GPS integrated navigation scheme with loose mode and indirect feedback correction is ascertained to satisfy the requirements of the autonomous vehicle's long-time and accurate navigation. Two filter models of the SINS/GPS integrated navigation of the autonomous vehicle, Kalman filter model and H∞filter model, are set up which are both based on the combination of position, velocity and attitude. Simulation results show that the autonomous vehicle's SINS/GPS integrated navigation based on Kalman filter has better performances in precision and real time when the system statistical properties are known, but it is sensitive to the change of disturbance; the SINS/GPS integrated navigation based on H∞filter is more adaptive and robust when the system statistical properties vary, but its real-time performance is inferior.(4) To improve the fault-tolerant ability and adaptability of the navigation system, a concentrated integrated navigation method of the autonomous vehicle in which SINS is aided by many sensors including GPS, machine vision, digital map and speedometer is proposed. The observation model of this navigation method is deduced in detail and set up. To diagnose the sensor's fault more conveniently, an adaptive federated filter method of the autonomous vehicle's integrated navigation based on SINS/GPS/machine vision/digital map/speedometer is also proposed. Simulation results show that both multi-sensor integrated navigation methods can provide accurate and reliable navigation information even when GPS is continuously interrupted...