Implementation of a ROS-based Autonomous Multi-Agent Cooperative SLAM Approach

Nowadays, mobile robots are utilized across a multitude of domains, including everyday life and research and work environments. Consequently, research in the field of autonomous robots has expanded considerably over the past few decades. In order to perform a task, a mobile robot must be able to localize itself in the surrounding environment. However, the operating area of a mobile robot may often be unknown, inaccessible, or dangerous. In such cases, Simultaneous Localization and Mapping (SLAM) is a key functionality that allows mobile robots to construct a real-time map of an unknown environment, while simultaneously localizing themselves within it. Due to its significance, SLAM conducted by a single agent in an isolated setting has long been regarded as the state of the art in mobile robotics research. However, in complex scenarios, single-agent SLAM is susceptible to limitations. To achieve faster and more accurate results, the Multi-Agent Cooperative SLAM (C-SLAM) system was born. The motivation behind the expansion of the research domain is rooted in the increasing demand for more efficient, scalable robotic systems capable of autonomous exploration in dynamic and complex environments. C-SLAM extends traditional SLAM by allowing multiple robots to collaborate in map construction and simultaneous localization. Another important step in SLAM research is Active SLAM. It enables robots to actively plan their movements and explore unknown environments, reducing map uncertainty through strategic decision-making, without active human assistance. A further focus of this research is the integration of Active SLAM (A-SLAM) into the collaborative framework, leading to Active Collaborative SLAM. With the advent of artificial intelligence, machine learning and robot learning, the Active SLAM will be a key area of robotic research in the future. The initial section of the thesis presents a comprehensive overview of existing SLAM techniques in both single-robot and multi-robot scenarios, highlighting the limitations of single-robot systems and the opportunities presented by multi-agent architectures. The second part of this thesis work introduces an autonomous centralized multi-agent approach, implemented using the Robot Operating System (ROS2 Humble) framework. Its performance is validated through both simulation (the Gazebo environment) and real-world experiments on two TurtleBot3 Burger platforms. In the considered system, each agent moves autonomously and creates a local real-time map independently. At the same time, a central server merges the two maps into a global, more accurate map. The single-agent SLAM algorithm used is the SLAM Toolbox. The path is planned locally on the single agent by the global planner A*, then smoothed by the B-spline and executed by the Pure Pursuit algorithm.