Pose Graph Based Simultaneous Localization And Mapping For Multi-Robot

Learning maps under the pose uncertainty is often referred to as the simultaneous localization and mapping(SLAM). With the native features of robustness and high efficiency, multi-robot based SLAM algorithm has show its superiorities to building map and locating itself. With insights into the structure of the SLAM problem and advancements in the fields of sparse linear algebra, pose-graph base apporaches can solve the full SLAM problem very well.This paper mainly research on the pose-graph based multi-robot SLAM problem. Our approach has been verified by performing experiments on Turtlebots. Firstly, robots need to building pose graph independently. Post graph consists nodes and edges. Nodes represent the estimation of the robot pose and the edges indicate the transformation between two nodes. Meanwhile, each robot should also detect other robot which comes into ite field of vision(FOV) to construct link-node. On the contrary of using special marks to help robots detecting each other, we combine the cascade Ada Boost classifier with gray histogram matching to improve the detecting accuracy. Robot A will transform its latest node together with its sensor data to the robot B when it is detected. Robot B will use this sensor data to construct a node and compare it with previous nodes to compute the transformation. If successful, this node will be add to pose graph, and we call this special node as link-node. At last, robot upload its pose graph to server when it finish its map building task. The link-node can be a bridge to merge maps constructed by individual robot. As the link-node exists in both detecting and detected robot’s pose graph, therefore we can convert a robot’s pose graph to the other ’s coordinate frame. When we optimize the merged pose graph, we also obtain the 3-D dense color map of the entire environment.