Payá Luis, Reinoso Oscar, Jiménez Luis M, Juliá Miguel
Department of Systems Engineering and Automation, Miguel Hernandez University, Elche, Alicante, Spain.
Q-Bot Limited, Block G, Riverside Business Centre, Bendon Valley, London, United Kingdom.
PLoS One. 2017 May 2;12(5):e0175938. doi: 10.1371/journal.pone.0175938. eCollection 2017.
Along the past years, mobile robots have proliferated both in domestic and in industrial environments to solve some tasks such as cleaning, assistance, or material transportation. One of their advantages is the ability to operate in wide areas without the necessity of introducing changes into the existing infrastructure. Thanks to the sensors they may be equipped with and their processing systems, mobile robots constitute a versatile alternative to solve a wide range of applications. When designing the control system of a mobile robot so that it carries out a task autonomously in an unknown environment, it is expected to take decisions about its localization in the environment and about the trajectory that it has to follow in order to arrive to the target points. More concisely, the robot has to find a relatively good solution to two crucial problems: building a model of the environment, and estimating the position of the robot within this model. In this work, we propose a framework to solve these problems using only visual information. The mobile robot is equipped with a catadioptric vision sensor that provides omnidirectional images from the environment. First, the robot goes along the trajectories to include in the model and uses the visual information captured to build this model. After that, the robot is able to estimate its position and orientation with respect to the trajectory. Among the possible approaches to solve these problems, global appearance techniques are used in this work. They have emerged recently as a robust and efficient alternative compared to landmark extraction techniques. A global description method based on Radon Transform is used to design mapping and localization algorithms and a set of images captured by a mobile robot in a real environment, under realistic operation conditions, is used to test the performance of these algorithms.
在过去几年中，移动机器人在家庭和工业环境中大量涌现，以解决诸如清洁、协助或物料运输等任务。它们的优点之一是能够在广阔区域运行，而无需对现有基础设施进行改造。得益于其配备的传感器及其处理系统，移动机器人成为解决广泛应用的通用选择。在设计移动机器人的控制系统，使其在未知环境中自主执行任务时，期望它能对自身在环境中的定位以及到达目标点所需遵循的轨迹做出决策。更简洁地说，机器人必须找到两个关键问题的相对较好的解决方案：构建环境模型，以及估计机器人在该模型中的位置。在这项工作中，我们提出了一个仅使用视觉信息来解决这些问题的框架。移动机器人配备了一个折反射视觉传感器，可提供来自环境的全向图像。首先，机器人沿着要纳入模型的轨迹行进，并利用捕获的视觉信息构建该模型。之后，机器人能够估计其相对于轨迹的位置和方向。在解决这些问题的可能方法中，本工作采用了全局外观技术。与地标提取技术相比，它们最近已成为一种强大而高效的替代方法。基于拉东变换的全局描述方法用于设计映射和定位算法，并使用移动机器人在实际运行条件下于真实环境中捕获的一组图像来测试这些算法的性能。
