Research On Simultaneous Localization And Mapping Of Robot Based On Line Segments

Simultaneous Localization and Mapping(SLAM)becomes a hot and challenging research topic in robotics in recent years.It is a basic and fundamental issue in robotics that has been developed with the integration of probabilistic theory,robot kinematics and sensor technology.The main research contents of SLAM problem include the perceiving of environmental information with sensors,constructing map model of unknown environment,updating of the robot pose with environment information to realize precisely positioning.For service robots,the accuracy of satellite positioning systems such as GPS and COMPASS is difficult to meet the navigation and positioning requirements in indoor environments.Moreover,due to the existence of random errors and systematic errors in the inertial navigation system on the robot,the long-term operation often causes excessive posture deviation.Therefore,the robot can not perceive its real pose with its own inertial navigation system,as if it is kidnapped.The key of SLAM is to use environmental information to estimate the real pose of the robot and to rescue the robot from the state of kidnapping with more precise positioning.At the same time,the precise positioning of the robot pose improves the accurate description of the constructed map for the unknown environment.The main contents in this dissertation are described as follows.A method for extracting topological features such as line and corner from laser scanning data to meet the needs of mobile robot positioning in a structured environment is proposed in this paper.Firstly,the map environment is scanned by a laser sensor and the environment information is saved in the form of points.Then,the straight-line segments are further extracted from these points by iterative line extraction algorithm and the least squares fitting method.Besides the linear features,there are many corner structures generated by two adjacent walls.The task of many mobile robots SLAM can be solved by corner mapping.The method first needs to extract feature points,and then extract the straight-line segments.Afterwards,The corner point is obtained by calculation of the intersection point of the two line segments.Finally,a matching algorithm is designed based on the distance and angle offset.So far,the the preliminary data for positioning and mapping are well prepared.In this paper,corners are extracted as landmarks of Extended Kalman Filter(EKF)algorithm by feature detection,and corners are taken as landmarks.The feature of the corner feature is also the feature of the straight line segment.EKF-SLAM method is used in robot localization and mapping.Firstly,the Gaussian noise assumption is adopted for the motion and prediction of the robot,so that the sum of the uncertainty of the posteriori error is relatively small.Then the map landmark information and the pose of the robot are described by the same state vector,and the pose of the robot is pre-predicted by inertial navigation.It is estimated that this state vector and Covariance matrix between robot pose and landmark are updated by the EKF method.The experimental results show that the proposed method can be applied to maps with corner points as the basic features with good robustness.Finally,the pose of the robot and the map information represented by the corner features are obtained.Since the straight line segments extracted in each cycle are relatively short,the straight line segment matching operation is performed between the small features.There are a large number of short line segments in global map.If all the straight line segments as the landmark of the straight line EKF-SLAM are stored in the vector state of the robot,the efficiency of the algorithm will be greatly affected.In this paper,in the premise of observing the straight line of the linear observation model,the Mahalanobis distance is introduced to simplify the matched straight lines.A method of transforming a straight line into a point in the Hough domain can filter the line features on the basis of the line extraction matching association and reduce the complexity of the EKF algorithm.Due to the large number of features of the EKF-SLAM algorithm for linear features,the generation of map features is more complete.The parameters of the robot's EKF-SLAM are derived under the same motion model with corner features.In the process of mapping,the method of feature merging is proposed for two common map feature relationships,and the map is optimized apparently.Finally,robot localization and mapping based on the linear features are realized.The EKF-SLAM algorithm is implemented by C++ programming using corner points and straight lines as map features.The PDX3 is carried out in MobileSim soft.Experiment task has been well accomplished.