Formation Navigation and Relative Localisation of Multi-Robot Systems

Abstract

When proceeding from single to multiple robots, cooperative action is one of the most relevant topics. The domain of robotic security systems contains typical applications for a multi-robot system (MRS). Possible scenarios are safety and security issues on airports, harbours, large industry plants or museums. Additionally, the field of environmental supervision is an up-coming issue. Inherent to these applications is the need for an organised and coordinated navigation of the robots, and a vital prerequisite for any coordinated movements is a good localisation.<br /> This dissertation will present novel approaches to the problems of formation navigation and relative localisation with multiple ground-based mobile robots. It also looks into the question what kind of metric is applicable for multi-robot navigation problems. Thereby, the focus of this work will be on aspects of <br /> 1. coordinated navigation and movement<br /> A new potential-field-based approach to formation navigation is presented. In contradiction to classical potential-field-based formation approaches, the proposed method also uses the orientation between neighbours in the formation. Consequently, each robot has a designated position within the formation. Therefore, the new method is called directed potential field approach. <br /> Extensive experiments prove that the method is capable of generating all kinds of formation shapes, even in the presence of dense obstacles. All tests have been conducted with simulated and real robots and successfully guided the robot formation through environments with varying obstacle configurations. In comparison, the nondirected potential field approach turns out to be unstable regarding the positions of the robots within formations. The robots strive to switch their positions, e.g. when passing through narrow passages. Under such conditions the directed approach shows a preferable behaviour, called “breathing”. The formation shrinks or inflates depending on the obstacle situation while trying to maintain its shape and keep the robots at their desired positions inside the formation. <br /> For a more particular comparison of formation algorithms it is important to have measures that allow a meaningful evaluation of the experimental data. For this purpose a new formation metric is developed. If there are many obstacles, the formation error must be scaled down to be comparable to an empty environment where the error would be small. Assuming that the environment is unknown and possibly non-static, only actual sensor information can be used for these calculations. We developed a special weighting factor, which is inverse proportional to the “density” of obstacles and which turns out to model the influence of the environment adequately. <br /> 2. relative localisation <br /> A new method for relative localisation between the members of a robot group is introduced. This relative localisation approach uses mutual sensor observations to localise the robots with respect to other objects – without having an environment model. Techniques like the Extended Kalman Filter (EKF) have proven to be powerful tools in the field of single robot applications. This work presents extensions to these algorithms with respect to the use in MRS. These aspects are investigated and combined under the topic of improving and stabilising the performance of the localisation and navigation process. Most of the common localisation approaches use maps and/or landmarks with the intention of generating a globally consistent world-coordinate system for the robot group. The aim of the here presented relative localisation approach, on the other hand, is to maintain only relative positioning between the robots. <br /> The presented method enables a group of mobile robots to start at an unknown location in an unknown environment and then to incrementally estimate their own positions and the relative locations of the other robots using only sensor information. The result is a robust, fast and precise approach, which does not need any preconditions or special assumptions about the environment. To validate the approach extensive tests with both, real and simulated, robots have been conducted. For a more specific evaluation, the Mean Localisation Error (MLE) is introduced. The conducted experiments include a comparison between the proposed Extended Kalman Filter and a standard SLAM-based approach. The developed method robustly delivered an accuracy better than 2 cm and performed at least as well as the SLAM approach. The algorithm coped with scattered groups of robots while moving on arbitrarily shaped paths. <br /> In summary, this thesis presents novel approaches to the field of coordinated navigation in multi-robot systems. The results facilitate cooperative movements of robot groups as well as relative localisation among the group members. In addition, a solid foundation for a non-environment related metric for formation navigation is introduced.