Research On Cabin Indoor Robot Mapping And Exploratory Localization Technology

Mapping and exploratory localization are the key steps in the actual deployment of a robot,which is the basis of the robot autonomous navigation.The narrow monotonous or low quality illumination environment leading to performance degradation of visual SLAM and laser SLAM is defined as the cabin indoor environment.Aiming at the problem that the robot cannot accurately complete the localization and mapping in the cabin indoor environment,a mapping and exploratory localization method based on multi-sensor fusion is designed,and a system is completely constructed by sensor information preprocessing,front-end mapping with pose tracking and back-end optimization.Then high-precision probability grid maps and robot trajectories can be obtained.Firstly,the modeling of monocular vision camera and single line lidar is carried out respectively.The pose tracking and local mapping methods based on the optimization method can be studied.The fundamental reasons of the faulty mapping and localization by single sensor in the cabin indoor environment are discussed.For visual SLAM,the low texture and low quality illumination will lead to decrease the number of stable tracking corners points.For lidar SLAM,similar scan information makes matching result wrong,which leads to wrong pose tracking and map insertion process.The IMU is modeled to complete the derivation of preintegration for a series of IMU information sequences.Secondly,calibrating the sensor,the consistency between the estimation and optimization methods of robot pose and map probability is proved.In order to control the scale in the process of robot mapping and exploratory localization,a sliding window is designed on the basis of tightly coupled information fusion,and an optimization function is constructed.In order to improve the robustness of the visual optimization item in the slide window,the number of tracking corners points and the movement state of the robot are evaluated,the weight of visual information in the slide window is adjusted by the information matrix.The volatilization of error term is restrained by robust kernel function.The effectiveness of visual scoring was verified by a practical scene experiment.Thirdly,a backend based on loop closure and pose graph optimization is designed to suppress the accumulated errors in mapping and exploratory localization.Image feature signature is used to perform visual loop closure,and its effectiveness is verified by dataset experiment and practical experiment.Laser loop closure is realized by retrieving variable resolution global probability grid map.Pose graph optimization is introduced to strengthen the constraint relationship between visual keyframes.Finally,a complete cabin indoor robot mapping and exploratory localization system is constructed.Corridor environment experiment,simulating aircraft cabin environment experiment and large scale sense experiments involving complex lighting have been implemented.Experimental results show that the proposed method can accurately construct a probabilistic gird map with a resolution of 5cm and provide robot trajectory information.