Research On Robust Mapping Methods In Unstructured Environments

Mapping in unconstructed environments is a fundamental and unsolved problem. In order to improve the robustness of mapping systems in unstructured environments, the thesis discusses and studies several aspects of the SLAM (Simultaneous localization and mapping) problem.Disambiguating environments is the key to improve the robustness. The thesis proposes a novel Harris corner detector based algorithm aiming at extracting highly stable and repeatable features from laser scans in unstructured environments with almost no prior knowledge. Comparing with traditional methods, the proposed method can be generally applied to all environment and achieve much richer features. And, at the same time, the detected features have attributes. Attributes based selection to features can increase robustness and add almost no extra burdens to mapping systems.The robustness of the data association step decides the robustness of the mapping system. The thesis proposes a decoupled joint compatibility test data association method. Noticing the independence between range and angle observation and the monotone increase of pose uncertainty in time update steps, and through introducing the innovation between estimated pose and the matched pose, the proposed method decoupled the data association problem. Therefore, the proposed method decreases the computation complexity of data association algorithms from O(pn2) to O(d2p) without impair the robustness, where p, nand d denote the amount of hypotheses, the amount of landmarks and the dimensionality of the environment, respectively.To decrease the influence came from dynamics of environments, the thesis proposes a novel Hopfield Neural Network based SLAM solution. In stead of apparently distinguish dynamic objects, the proposed method adopts a biologically inspired strategy:enforced memory and gradually forgetting. The proposed method eliminates traditional requirements on the velocity and amount of dynamic objects, improves the ability to distinguish environment, and increases the consistence of constructed maps. Consequentially, the robustness of the mapping system is increased.To deal with impact from unpredictable influences, like robot kidnapping, the thesis also proposes a loosely coupled framework that used wireless sensor network (WSN) nodes as an auxiliary to SLAM system. In the application contexts that the robustness is highly desired and global localization systems are not available; the robot dynamically sets up and localizes WSN nodes while it is mapping the environment. The nodes actually segment the environment, so, the pose can be improved, and the robustness is also improved.The proposed methods are tested in simulated environments, real environments and classical data sets to validate them and testify the efficiency. These methods can be applied individually to aspects of SLAM problem, and, at the same time, they can be used to construct a completed mapping system.