Research On Lane Level Navigation For Autonomous Vehicle

Automatic vehicle control systems which is designed with understanding of and accuratemodeling vehicle characteristics can be used to improve vehicle safety, fuel economy, ridingcomfort and travel efficiency of Driver-Vehicle-Road closed loop system. The handlingreliability of driverless car is much higher than human drivers’. The active role played bydriverless car in regular traffic makes related technologies get a lot of attentions.In completely unfamiliar or wild environment, driverless car totally depends on itsperception equipment to model the surrounding environment. With the environment’s model,driverless car searches for feasible routes and performs motion planning and motion controlto achieve the mission objectives. The kind of driverless car which is driving on the structuralroad is able to make use of the map of traffic environment to reduce the dependence on high-precision perception equipment. The objective of this thesis is to explore the application ofdigital maps on driverless car and to conduct research on navigation systems for driverless car.Compared to the traditional car navigation systems, navigation systems for driverless carneeds solution with higher precision and certainty. So, in this thesis, lane-level navigation isstudied. The main content of this thesis are lane-level route planning and vehicle positioningwith multi-sensor fusion.Firstly, research on route planning for autonomous vehicle. According to the requirementof autonomous vehicle’s navigation system and the current situation of road traffic, themeaningofautonomousvehicle’snavigationsystemwasdefined.Theabstractmethodoflane-level navigation map for autonomous vehicle was proposed based on autonomous vehiclematches. The existing road resistance models were discussed and the determination of roadresistances with different optimization objectives were analyzed. According to the applicationrequirement and characteristic of different models, BPR model and Webster model were usedto calibrate the road resistance. The framework of lane-level route planning based on dynamicroad resistance was proposed. The common graph search algorithms and planning algorithmswere discussed and A*algorithm was chosen to conduct route planning.After that, probe into vehicle positioning with multi-sensor fusion. The framework offusion methods to combine GPS, IMU and wheel speed with Kalman Filter was set up. Fulltime vehicle position estimation was achieved. Especially, when the GPS is disable, wheel speed is used to reduce the accumulative error from IMU. Methods from Vehicle Dynamicswere used to calculate the vehicle position with IMU and high precision was achieved. In DeadReckoning methods, vehicle trajectory was described as circular arcs and the change of vehicleposition was estimated with travel distance and the change of yaw angle.Finally, the main contents of autonomous vehicle navigation systems were realized andcertificated. The results show that the abstract method of lane-level map is able to describeLane-level topology map and form interconnecting lane network, dynamic route planningalgorithm is able to plan feasible routes for different optimization objectives and routeplanning for shortest travel distance, shortest static travel time and shortest dynamic traveltime were achieved. Full-time and high-frequency vehicle positioning was achieved withfusion methods to combine GPS, IMU and wheel speed. Simulations and experiments onfusion methods with/without Dead Reckoning were conducted. The results show that methodswith Dead Reckoning has much better effects.