A Research On The Collaborative Simultaneous Localization And Dense Map Construction Method

Collaborative SLAM systems can improve the speed of gathering information about unknown environments.Currently,most multi-robot collaborative map reconstruction methods build unstructured and sparse maps,which can only be used for localization but not directly for navigation and obstacle avoidance applications.To address these problems,this paper aims to design a multi-robot collaborative simultaneous localization and dense map construction system,propose the robustness algorithm of multi-robot pose map fusion,and explore the construction method of globally consistent dense maps,so as to provide technical support for the functional modules of collaborative detection and path planning of multi-robot systems in unknown scenarios.The main research contents are as follows:(1)A position-corrected stereo visual inertial odometry using land marks’ bias is proposed.Although the stereo camera can measure the depth of land marks,its measurement accuracy is easily affected by the image matching accuracy and parallax angle.This paper derives the bias information equation in the extraction of 3D land marks by the stereo camera,which reveals the biased nature of the positional estimation error,and investigates a method to correct the positional estimation by using the inverse depth information of 3D land mark.Experiments show that the RMSE metric of absolute trajectory error is improved by 12.22% on average compared with the VINS-Mono algorithm.(2)A false positive loop closure rejection method based on two-step consistency checking is proposed.False positive loop closures in collaborative SLAM systems can seriously interfere with the positioning accuracy of SLAM systems.To address this problem,a two-step consistency check method is proposed,with the first step error rejection targeting false positive loop closures checked in a single robot and the second step rejection of loop closures measurements between robots that do not pass the consistency check based on a pairwise consistency check algorithm with maximum consistency clusters.(3)A collaborative dense map construction algorithm based on a centralized architecture is investigated.A global keyframe database is constructed on the central server to eliminate false positive loops,and anchor nodes is used to unify the coordinates of multiple robots into a global coordinate system,and an optimised model for global pose fusion estimation is constructed.The Marching Cubes algorithm was used to build a collaborative dense map of the scene.The experimental results show accurate localization results compared to those of SLAM systems based on centralized architectures that can only build global sparse map.