Analysis, Architecture, and Fusion Methods for Vehicle Automation
Albores Borja, Carlos
MetadataShow full item record
Autonomous Vehicles (AVs) are automated vehicles to carry out a specific task, without direct human intervention. AVs are mobile robotic applications that have generated great interest in recent years due to their capacity to perform repetitive tasks in remote or harmful environments with extreme operating conditions. The applications and tasks of these devices vary from the transportation of material, to the exploration of planet's surfaces. AVs consist of selecting a vehicle originally designed for human drive, and installing the necessary components and systems to carry out the required tasks with autonomy. A methodology for converting a commercial vehicle into an AV with the proposed architecture is also introduced. To organize and control all the elements and functions of an AV it is indispensable the design of a physical and logical structure of these elements, known as an architecture. The architecture also specifies how the elements are coordinated and how they interact between themselves. This research introduces a control architecture for autonomous vehicles. This architecture makes an abstraction of the functions of the vehicle, with an emphasis on the vehicle's kinematic model. The architecture is modular and is structured mainly in a hierarchical way, with some modules of reactive behavior. One of the main elements of this vehicle's architecture, and of AVs researches in general, is the vehicle's position estimation function. Better state estimations results in better and more reliable performance. This research present a method to obtain a expression for the uncertainty in the odometry position estimate of a mobile vehicle using a covariance matrix whose form is derived from the kinematic model. We then particularize for a non-holonomic Ackerman driving type autonomous vehicle. However, obtaining a expression for the cross-covariance terms between the previous position of the robot and its actual increment of position is not straight forward. Thus, a formulation to obtain a expression for these terms is developed. Finally, special care must be taken into account when data for multiple sensors are fused, since it is easy to over estimate the state's precision using fusion techniques that do not consider correlation between the sensors' measurements. For this reason, techniques considering correlation such as probabilistic approaches or the covariance intersect algorithm are considered in the hierarchical data fusion scheme introduced in this thesis. To validate the aforementioned elements, an utilitarian carrier designed to be used in open space mining industry was automated, and trajectory following experiments were performed and analyzed. The vehicle was able to follow a desired path within an error of less than 1 meter using all the available sensors data.