Research On Cognitive Map Hierarchical Model For Autonomous Land Vehicle Navigation

Robotic mapping has been a highly active research area in artificial intelligence and robotics recently. Under the principle of biological navigation, in a view of environment cognition, map building technique has been researched for Autonomous Land Vehicle (ALV) navigation in this thesis, and hierarchical topology-action-metric map has been constructed in the complex outdoor environment for our project. Our research works include:(1) By analyzing and summarizing the existing cognitive map models, a new hierarchical topology-action-metric model is proposed which considered the character and navigation requirement of ALV. Topological map is used for expressing the global space to ensure global consistency, and geometric metric map is used for local space model to make the description of environment being simple and accurate. This hierarchical model express the environment information of ALV navigation reasonably and improve the efficiency of path planning.(2) In the realization of topological layer, firstly, the passable region is extracted from monocular vision by the marker-controlled watershed algorithm, then multi-frame images is jointed probably to generate the global consistent map, finally the road centerline is extracted from global consistent map by thinning algorithm, then the topological layer map has been established by encoding.(3) In the realization of action layer, the character of ALV is considered, and on the basis of mathematical description of vehicle kinematics and dynamics constraint, a new method to generate action layer map by building optimal path band is proposed. The path generated from the action layer map meet the constraints of vehicle kinematics and dynamics successfully.(4) In the realization of metric layer, the geometric map representation is the most important work. Ellipse clustering is used to process the LARAR data to generate the metric layer map. The planning has been improved by the generated metric map.Experiments show that the topology-action-metric hierarchical model performing very well in ALV real-time navigation.