Path planning for autonomous vehicles
In this study, we first address the problem of visibility-based target tracking for a team of mobile observers trying to track a team of mobile targets. Initially, we introduce the notion of pursuit fields for a single observer to track a single target around a corner based on the previous work. Pursuit fields are used to generate navigation strategies for a single observer. In order to account for the scenario with the presence of more than one observer or target, we propose a hierarchical approach. At first a ranking and aggregation technique is used for allocating each observer to a target. Subsequently, each observer computes its navigation strategy based on the results of the single observer-single target problem, thereby, decomposing a large multi-agent problem into numerous 2-agent problems. Based on the aforementioned analysis, we present a scalable algorithm that can accommodate an arbitrary number of observers and targets. The performance of this algorithm is evaluated based on simulation and implementation. To implement the strategy in reality, we further propose a setup of omni-directional camera, which can be used to get the visual information of the surroundings. With the help of this setup, we apply a position estimation technique for the pursuer to locate the evader. Experimental results show that the error has considerable effect when the measuring distance is very large. Due to this reason, the aforementioned tracking strategy is modified to keep the evader in an effective range for estimation. Finally, based on the error in position estimation, we present PID controllers for the pursuer to track the evader along a straight line. The responses of the proposed controllers are given by simulations.
Considering the situation that pursuer does not have an on board vision sensor, we propose a novel tracking strategy based on the information on social network. We first introduce the notion of common agents, who take pictures around and share them on social network website. In order to take advantage of these images, a network evolution algorithm and an image scanning algorithm are presented. Based on the information from these images, evader can be located accordingly. Implementation results are presented to validate the feasibility.
In the rest of the thesis, we address the scheduling and motion planning problem for an autonomous grain cart serving multiple combines. In the first part, we present the mathematical models of both combine harvester and grain cart. Based on the models, we propose a scheduling scheme which allows grain cart to unload all the combines without interruption in the harvesting activity. The proposed scheme is generalized to an arbitrary number of combines. In the second part, we present path planning analysis for the grain cart to switch between two combines. A numerical approach and a primitive-based approach are considered to obtain the time-optimal solution. The former approach needs a value function corresponding to the goal position to be computed beforehand. Based on the value function, a time-optimal path can be obtained accordingly. In the latter approach, path consists of singular primitives and regular primitives which ensure local time optimality. Finally, simulation results are presented to validate the feasibility of the proposed techniques.