Slam Based On Local Sub-graph Matching Solution

Mobile robot is a comprehensive system which includes the environment sensing, dynamic decision-making,action control and execution.Simultaneous localization and map building(SLAM) is a very important issue and basis in mobile robot field. The solving of SLAM problem make robot work well without the priori knowledge of the environment and other absolute localization system,make the absolute autonomy of the robot possible,and bring about broad prospects of the application of mobile robot.This paper focuses on the SLAM problem in unknown environment and it contents the following parts:First,the history and present study on mobile robot is introduced,then the most popular method for SLAM is addressed and analyzed.For the "kidnapped" robot problem to which most current SLAM(simultaneous localization and map building) approaches are invalid,a new solution based on local sub-graph matching is proposed which is called LSGM method.This approach improves the current architecture to SLAM problem,provides a new feature association algorithm based on optimized vertex matching,and incorporates the singular value decomposition method into robot localization.Finally,improved LSGM method is proposed and that is more robust in different situation.The main contribution of this paper is as follows:a.A new architecture for SLAM is proposed.After the line extraction,the landmarks in local map are computed and matched with the ones in global map,then the associated landmarks are found out and according to these,the pose of robot is estimated.There is no odometry information within sub-graph association algorithm and the other prediction pose from odometry is calculated independently.Thus avoid the system error which is always occurred in general SLAM method because of the non-linear prediction and update model,which is the main obstacle for "robot kidnap" problem.b.A new feature matching method is proposed.The landmarks extraction algorithm which based on vertex of line segments and the matching algorithm,that make it more robust for replacing nature landmarks with vertex.It is essential an N-P problem,and this paper propose a optimal method.c.Introducing SVD to robot localization.The method based on SVD is not recursive,and the accuracy and speed is better than normal approach.d.An adaptive threshold breaking points detect algorithm is proposed.This method concerns the relationships of sampling angle and sampling distance,enhance the veracity.In the end,this approach and other approach for SLAM are compared and the discussion are made in the term of the feasibility and the effectiveness of this approach based on local sub-graph matching method for the "kidnapped" robot problem in structured environment.