Research On SLAM Method For Multisource Information Fusion Of Mine Search And Rescue Robot

The intelligent and unmanned development of the coal industry is the key to achieving high-quality development.The difficulty and danger of post disaster rescue in coal mines are high.The use of mining search and rescue robots to replace rescue team members for environmental detection,information transmission,and rescue is an effective way to achieve safe and efficient rescue in coal mines.Due to the complex and harsh conditions of the post disaster underground environment,it poses great challenges to the robot’s environmental modeling and self positioning.This article aims to study laser,vision,and The synchronous positioning and map building(SLAM)method for inertial and other multi-source sensor information fusion is used to achieve efficient modeling and precise positioning of mining search and rescue robots in complex underground environments.The main research content is as follows:(1)Based on the analysis of the characteristics of the underground environment in coal mines,a testing platform for mining search and rescue robots is developed.Firstly,the characteristics of the underground environment are analyzed,and relevant requirements for explosion-proof,obstacle crossing,communication,and other aspects of mining search and rescue robots are proposed.Furthermore,the explosion-proof mechanical structure and hardware control framework of the robot are designed,and the SLAM software development framework and low-level motion control logic of the robot are constructed.This provides a practical experimental platform for subsequent theoretical and experimental research on algorithms(2)To address the issue of decreased accuracy and robustness of laser SLAM under conditions such as bumpy underground roads and rapid robot rotation,a precise positioning and map construction method for mining search and rescue robots based on laser inertia tightly coupled SLAM is studied.On the basis of Li DAR feature extraction,IMU is used to distort the features,and fast neighborhood search(IKD＿Tree)is used to calculate the point line and point surface residuals of the extracted features.Iterative Kalman filtering is tightly coupled with IMU to construct a 3D point cloud map with line surface features and obtain the radar pose at the same time.Field experiments were conducted in simulated underground tunnels,and the average positioning error was3.49%,indicating the effectiveness of the proposed method for conditions such as rapid rotation and turbulence.(3)To address the degradation of laser SLAM in underground environments with few structures,as well as the decrease in visual SLAM accuracy and robustness in low illumination and weak texture environments,a precise positioning and map construction method for mining search and rescue robots based on vision inertia tightly coupled SLAM is studied.Based on the actual working conditions underground,a research on optical flow tracking methods is proposed to use sliding window modules to reduce the impact of non single point light sources on tracking performance,while increasing the number of key frames to supplement front-end visual information,thereby constructing stable and reliable visual feature residuals.Study the principle of IMU pre integration and construct its residuals,and implement visual inertial tight coupling based on factor graphs.Field tests were conducted on simulated tunnels underground,with a positioning error of 2.49% and a mapping error of 2.91%.This indicates that the proposed method has certain adaptability for underground environments with few structures,low illumination,and weak textures.(4)To further improve the accuracy and robustness of positioning and mapping,a precise positioning and map construction method for mining search and rescue robots based on laser vision inertia tightly coupled SLAM is studied.In response to the degradation problem in the laser inertial SLAM experiment,a dual system is established to work with the algorithm failure problem in the visual inertial SLAM experiment in a light free environment,namely the LIO(laser inertial)subsystem and the VIO(visual inertial)subsystem.The two subsystems are tightly coupled and have multiple loops.When the algorithm operates in a dark environment,the VIO subsystem introduces radar pose as a prior initial value to reduce trajectory drift error.When the algorithm constructs images in tunnels with single geometric features,visual constraints are introduced into the LIO subsystem as a prior initial value for radar pose.At this time,the degree of trust in the VIO subsystem is determined by visual weights,and the relatively accurate camera pose is multiplied by a larger weight to give the LIO subsystem,optimizing the construction of radar pose.Finally,an experiment was conducted in a simulated underground tunnel,with a positioning error of 1.6% and a mapping error of 1.36%,indicating that the fused algorithm has higher positioning and mapping accuracy and is more suitable for post disaster environments in mines.The thesis has 76 figures,20 tables,and 111 references.