Research On Map Building For Multi-robotic System Based On Hierarchical Mixed Feature

In recent years, with the further research of robot navigation, the service taskdemand becomes more and more complex. To perform the complex tasks,multi-robotic system has been a hot topic for its obvious advantages such asparallelism, flexibility, fault tolerance and data redundancy. This thesis proposes atopology nodes matching algorithm based on mixed features and combines with localscan matching strategy to accomplish multi-robotic system map building task withoutknowing corresponding pose between each robot for large-scale unknownenvironment. Firstly, an improved SP2ATM is used to construct local maps formulti-robots. On this basis, under the hierarchical topological structure with mixedfeature, ICP algorithm is integrated with topological correlation to realize the mapmerging for multi-robotic system and correct the robot’s pose. To improve the robotautonomous recognition capacity, more area features are needed. So clusteringalgorithm is adapted for global topology region division. The main research works areas follows:1. Local layer construction based on mixed features.To improve robot environmental recognition capacity, a hierarchical mixedfeature (HMF) structure, which contains local feature layer, global topology layer andregion division layer, is constructed to serve for map building and region divisionalgorithm. For local feature layer, local scene is described by abstracting geometrydetailed features (GDF), Admissible Space Tree (AST) and Speed up Robust Features(SURF). Line segment algorithm and Grid-based shared Nearest Neighbor method(GNN) are introduced to abstract GDF. On this basis, the recognition capacity oftopology nodes can be improved by combining with SURF, and the feature matchingratio is used as prior information of map merging and region division algorithm.2. Improved SP2ATM algorithm based multi-robotic system map building.To accomplish robot path exploration and continuous map constructionincrementally for large scale unknown environment, an improved SP2ATM algorithmis utilized to construct scene topology structure. This algorithm constructs ahierarchical topology structure containing mixed feature such as SURF corner points,topology nodes and so on. Simultaneously, same area exploration between robots isavoided based on topology nodes updating rules to improve the multi-robot mapbuilding execution efficiency. Besides, to improve the topology nodes localizationaccuracy, Rao-Blackwellisation particle filter (RBPF) is introduced to correct robotpose.3. Map merging for multi-robotic system based on scan matching. A map merging method for multi-robotic system integrating vision feature withlocal scan matching is proposed without knowing the corresponding pose betweenrobots. Firstly, HMF is abstracted with the key-frame thought, and the sceneinformation is associated with topology nodes matching strategy to lower thematching computation complexity. Then, the corresponding matrix of local map iscalculated using Iterative Closet Point (ICP) algorithm, and pose estimation error iscorrected with the main-secondary robotic model. Simultaneously, to ensure theglobal consistency of map information, a feedback system is imported to solve themismatching problems. Finally, during the map merging process, the similaritybetween maps is regarded as measurement level, and the map merging task formulti-robot is accomplished by searching overlaps of each local map.4. Mobile robot region division based on spectral clustering algorithm.In order to obtain more region information, the theory of spectral clustering basedon Min-Ncut is introduced according to the undirected weighted graph generated bytopology map. To improve the clustering accuracy, the node matching informationbetween GDF and SURF is combined with node shortest path information to constructthe similarity matrix. Then the topology nodes are classified by spectral clusteringmethod to accomplish region division for mobile robot.The validity and practicability of the proposed map building approach formulti-robotic system is validated by a lot of experiments on USARSim simulationplatform and the mobile robot Pioneer3-DX platform. The experimental results have acertain theoretical and practical value to the research of multi-robotic system.