Research On Map Building Of Multiple Mobile Robots

Robotics has already been widely developed and used in many fields. Urged by investigation and application, research of multi-robot system has received more and more attention, and it is gradually becoming an active field full of challenge. In some tasks, such as military operation, aviation, services and RoboCup, map building is not only very important to accomplish autonomous navigation and other complex intelligent tasks, but also embodies perception ability and intelligence of multi-robot systems. So it is an important and key problem of research on multi-robot to building environment map accurately and effectively by multiple robots.This work focuses on the basic and important problem in multiple robots research—multi-robot map building. It addresses the problem of continuous obstacle avoidance and collision avoidance strategy, simultaneous localization and mapping (SLAM) based on relative observations and local map merging of a mobile robot team.First, the problem of multi-robot obstacle avoidance and collision avoidance is discussed. Aiming at sonar robots, a kind of direction selection rules based on the avoiding sub-behavior states is designed, to choose the right avoiding direction, with the cooperation of the current obstacle orientation been detected. For the collision avoidance strategy of multi robots, a model named "No-Traffic-Light-Crossing" is built. In this model, a Pass-Priority evaluating function is designed, to decide which of the robots in the model will gain the pass priority. Digraph is used to describe the avoidance relation of multi-robot system, and the "dead-lock" problem will be solved through stepwise eliminate the cycles of the digraph.Second, a novel approach to multi-robot simultanieous localization and mapping (SLAM) based on Extended Kalman Filter (EKF) is presented. The EKF-SLAM approach is improved by support vector machines. According to relativity of innovation, the measurement noise covariance was adjusted adaptively. The method could cope with divergence problem caused by the insufficiently knowing of the prior filter statistics and improve the accuracy of SLAM. Then, the improved EKF-SLAM approach used in a single robot in common was extended to a multi-robot system. Relative observations between robots when meeting was processed to execute coordinate transformation and map merging. In addition, by considering the uncertainty of sensor information, an improved matching of landmarks was introduced to the process of map fusion to increase accuracy of localization and mapping. Compared with typical ones, this approach can accomplish multi-robot simultaneous localization and mapping accurately and effectively without using any initial pose information of robots and no need for robots to explore duplicate areas, so as to be more suitable for a variety of complex cases in application with less restrictions.Aiming at multiple sonar robots, a multi-robot FastSLAM approach is discussed. First, the standard particle filter and particle swarm optimization algorithm are incorporated into the filtering framework of this approach. The newest observations are introduced to adjust particles' proposal distribution, so as to largely reduce the sample size necessary for localization and mapping and effectively relieve the particle degeneracy problem while ensuring the algorithm precision. In addition, considering that the typical resampling process always leads to the loss of diversity in particles, a probabilistic operator is introduced to keep the diversity of particle swarm. Second, the improved FastSLAM approach is extended to a multi-robot system. The relative observations when robots' meeting is used to initialize the particle filter, and the subsequent and prior observations from robots are combined into a global grid map by using virtual robots. Experimental results show the approach has high accuracy and stability as well as flexible map representations.Considering that the robots may never encounter at all in real environment, a multi-robot mapping approach based on grid matching is presented. The approach lets all robots operate individually and then tries to merge the different local grid maps into a single global one. Without using any pose information of robots, the process of map merging is performed by measuring the similarity between grid maps. Distance transforms and an improved genetic algorithm are used to effectively search the maximum overlap at which the local maps can be joined together. Experimental results show that this approach can accomplish multi-robot map building accurately and effectively without using any robots' position information, and be more suitable for sonar robots in applications for mapping with less restrictions.Finally, we summarize the general work of this thesis and give a short outlook on possible future research.