Research Of SLAM And Path Planning Algorithm Based On Sensor Fusion

Robot technology is a comprehensive technology representing the development of modern science and technology,including information technology,intelligent technology,mechanical automation technology and so on.With the evolution of social development technology,the research of artificial intelligence technology has become a hot spot nowadays.The SLAM(Simultaneous Localization and Mapping,SLAM)algorithm based on artificial intelligence machine vision uses stereo or RGBD camera to collect and analyze data similar to human eye,which provides effective input for robot more intelligent navigation and interaction.The algorithm of motion planning has a long history,and new solutions have been developed over the years to provide a rich reference for robot automation.These technologies are widely used in the fields of unmanned driving,intelligent manufacturing,UAV and so on.This thesis mainly studies the basic principle of the current SLAM algorithm,and puts forward the SLAM system of fusion depth vision and inertial navigation sensor,which improves the positioning accuracy compared with the single sensor SLAM system.The two-dimensional projection algorithm based on growth quadtree is also proposed.The A* path cost calculation method of adaptive SLAM frame output is designed,and the complete system of robot from location construction to robot path planning is completed.From the experimental point of view,this method improves the positioning accuracy and the efficiency of drawing building,and improves the navigation algorithm of the robot.The main contents of this thesis are as follows:1.Existing SLAM schemes without sensor fusion have limited error estimation accuracy and poor robustness.Exploiting IMU sensors in rapidly moving scenarios will have higher accuracy,and visual sensors cannot to get clear visual field images for calculation.However,in the case of relatively smooth and slow motion,the IMU sensor will produce data in the static state due to the disadvantage of drift,which increases the cumulative error of pose calculation.For better robustness and accuracy of the system,SLAM correlation of fusion inertial and visual sensors is studied in this thesis.For this subject,a new tight coupling strategy is designed,and the pose is optimized by sliding window method and beam adjustment method.After experimental verification,the accuracy and robustness of the scheme are improved.2.At present,the building part of simultaneous localization and map construction algorithm of visual robot mainly adopts 3D octree map.Although its map storage capacity is large,but the map cannot to be expanded in real time,and the common dynamic things in indoor scenes are also difficult to deal with because they ignore the large noise points.To this end,this thesis proposes a new real-time grid two-dimensional map construction method based on the growth-type quadtree structure,which reduces the dimension of three-dimensional voxel map to two-dimensional grid map,increases the prediction of dynamic feature point trajectory,and enriches the environmental information carried by navigation map without loss of 3D spatial information.The experiment of real indoor scene shows that this algorithm can display the location information of obstacle more accurately in the map,reduce the storage space of map significantly,and improve the speed of building map.3.The classical path planning algorithm A* algorithm cannot to get the global optimal solution sometimes because of the different setting of the cost evaluation function and the node search strategy.In the process of path planning,the point of broken line will also affect the navigation efficiency.Furthermore,the path planning algorithm suitable for the input of the SLAM algorithm has not been tightly coupled.To solve these problems,this thesis designs a node grid map construction method based on the new map abstraction principle,which is coupled with an improved A* path planning algorithm adapted to visual SLAM.Finally,the path is smoothed by three B spline curves.By this method,the efficiency of the robot during displacement is improved,and the action decision-making ability of the visual SLAM robot is enhanced.4.Firstly,the experimental part of this thesis verifies that the SLAM under inertial vision sensor fusion has better accuracy and robustness.Secondly,a two-dimensional map for ground robot navigation is established.Finally,the design of path planning and obstacle avoidance algorithm in the field of vision is completed,so that the robot can explore and map the environment in the unfamiliar indoor environment.This thesis forms a complete framework of robot navigation scheme,and proves the superiority of the algorithm through a large number of related experiments.