Multi-Robot Collaborative Exploration Method For Reconstruction Of Unknown Environment

Autonomous exploration is a method for robots to dynamically discover the surrounding environment and build maps in real time during the task process,which is suitable for surveying and mapping,disaster exploration and patrol inspection,etc.Especially,compared with single robot,multi-robot has more advantages in large-scale scenes.In the process of multi-robot collaborative exploration,each robot should identify its next exploration target point and coordinate with other robots,with the aim of reducing redundant information in the exploration process and covering an initially completely unknown environment in the shortest time.This paper focuses on how to evaluate the target points of robots effectively and how to realize efficient robot behavior planning in3 D scene,so as to realize rapid 3D reconstruction in unknown environment.And the contributions to this paper are given here:(1)In the reconstruction-oriented exploration problem,the exploration efficiency and reconstruction quality must be considered simultaneously when determining the next exploration target point.This paper proposed a multi-robot viewpoint evaluation method based on truncated signed distance function map(TSDF-Map).Based on the model of eyelet imaging,the environment is represented by TSDF-Map,and the value of viewpoints is estimated through current map.By defining the segmented info gain function,the effectiveness of unknown areas and the uncertainty of known areas are quantitatively calculated,and then combined with the robot motion cost,the total utility formula for evaluating viewpoints is obtained.This method realized the robot to explore the areas near obstacles and the areas with low reconstruction accuracy first,and improved the efficiency of unknown environment exploration and map quality for reconstruction.(2)To solve the problems of computationally expensive planning problem in 3D scenes,this paper proposed a multi-robot collaborative exploration method based on multiple Rapidly-exploring Random Trees(Multi-RRT).The method included two modules: exploration and coordination.Firstly,the local RRT expanded in the environment around the robot and the global one found frontiers all over the region.And the value of the robot's potential viewpoints would be caculated by the method in(1).Secondly,the coordination module used the method based on hungary algorithm to allocate candidate viewpoints.Robots used the obtained candidate viewpoints to expand their local RRT and generate exploration paths.Then,through the strategy based on utility maximization,the paths between robots were collision coordinated.This method realized efficient exploration of multi-robot in 3D scene.(3)The above methods were proved in the simulation environment named AirSim.The experimental platform was built with AirSim and an open-sourced framework named ROS.Then,the robot exploration and reconstruction experiments were carried out in a highly realistic environment.And the methods proposed in this paper were systematically analyzed and verified by exploring time,exploring coverage and accumulated information gain.