Research On Mobile Robot SLAM Algorithm Based On Multisensor Fusion

As a forward-looking technology for mobile robots and autonomous driving,simultaneous localization and mapping(SLAM)has always been a hot research direction.The system collects data from the surrounding environment and the robot itself through sensors,and estimates the pose of the mobile robot,thereby achieving functions such as robot positioning and map construction in unknown environments.However,the application scenarios of a single sensor positioning system have limitations,making it difficult to effectively complete the positioning and mapping tasks in complex environments.Therefore,this article proposes a localization algorithm based on multi-sensor fusion to achieve robot localization in complex environments.For environments with weak textures and few key points,the traditional visual SLAM algorithm odometer is difficult to accurately track key points,resulting in reduced robustness of the positioning mapping system and poor positioning accuracy of the robot.To address this issue,this article proposes a front-end odometer based on multi-sensor fusion by adding a wheel encoder.The wheel encoder and inertial measurement unit are used to estimate the position and pose of the robot to make up for the deficiency of the visual odometer in the weak texture environment,so as to enhance the positioning accuracy of the system.The effectiveness of the algorithm was verified through the KAIST urban dataset on the ROS platform.Compared with the VINS Mono algorithm,the absolute trajectory error of the odometer was effectively reduced,and the algorithm has high accuracy and reliability.When the mobile robot moves at a constant speed,the scale information of the inertial measurement unit of the SLAM system will be missing,and the observability of the system will be reduced,resulting in data drift and reduced positioning accuracy of the system.In response to this issue,this article adds a backend pose optimization module to the system,and together with a front-end odometer based on multi-sensor fusion,forms a multi-sensor fusion SLAM system.In the loop back section of the system backend module,the pose estimation problem of the robot is described as a nonlinear least squares optimization problem and solved through visual residual constraints,pre integration residual constraints,and planar motion constraints.Experimental results have shown that compared to the VINS Mono algorithm,the proposed SLAM algorithm based on multi-sensor fusion effectively reduces the absolute trajectory error.In the process of pose optimization of SLAM system,images are compared frame by frame through visual feature points to judge loopback frames,which will lead to increased overhead.In this regard,this paper improves the loop back part of the algorithm,adds wheel odometer data to the bag-of-words model that stores visual feature point data,and uses the distance constraint of wheel odometer to reduce the loop back frame decision frequency.