Research On Multi-robot Collaborative Autonomous Mapping Strategy Based On Visual Perception

The use of robots for 3D reconstruction of real scenes has a wide range of applications,such as disaster and rescue sites,planetary exploration and factory mapping.At present,there are a variety of SLAM methods to achieve this function,but most of them focus on humancontrolled robot localization and mapping,so that the final mapping effect is limited by the personal experience of the operator.This topic mainly studies the situation without human intervention,the multi-robot system in the completely unknown environment according to the perception of the vision sensor,the cooperative autonomous planning of their own moving trajectory,and finally complete the three-dimensional reconstruction of the entire environment.It involves visual SLAM method,dynamic object processing,boundary detection,perceptual task point extraction,multirobot collaboration and other key contents.The details are as follows:Firstly,the visual SLAM algorithm is studied in order to achieve a good mapping effect.On the one hand,in order to achieve the real-time effect,the location and mapping module were decoupled.On the basis of calculating the robot's position and pose by optical flow method,the dense mapping task was completed by using polar search and block matching technology.On the other hand,for the potential dynamic objects in the environment,deep learning is adopted to identify them,so as to achieve the effect of no reconstruction.In addition,in order to express the latest environmental state,the map is updated according to the observed values of the same position at different times through the representation form of octree map,so as to achieve the effect of adding or deleting obstacles.Secondly,the extraction method of task view is determined in order to realize robot autonomous map building.The fast searching random tree algorithm is used to detect the boundary,which ensures the speed and integrity of boundary detection.Then,considering the perception range of vision sensor,task views are extracted by gray center method,and the validity of each task view is analyzed.Then,the task assignment of multi-robot was carried out based on the theory of optimal mass transfer.Considering the degree of discretization assigned to each robot task subset,the cost of moving the robot to the task subset and the scanning ability of the robot,a discretization objective function is proposed,which ensures the minimum cost and the highest efficiency of robot mapping,and at the same time makes the load balance among robots.Finally,the above algorithms are integrated and evaluated.The overall framework of the system is determined,and multi-dimensional evaluation indexes are proposed to quantitatively evaluate the algorithm from the aspects of running time,overall energy consumption,load balance and reconstruction integrity.Different multi-robot collaborative autonomous mapping algorithms are used in different scenarios to verify the superiority and robustness of the algorithm.