Desenvolvimento de uma técnica de navegação autônoma para ambientes externos

Abstract

According to (Robotics IndustryAssociation, USA), mobile robotics is the ability of robots to move from one location to another transporting materials and providing services. This is one of the most promising branches of robotics, due to the prospects for high growth, with the insertion of applications in the most different areas. In this direction, this work aimed to carry out a new approach, developing an algorithm for mapping environments using color filters to detect navigable or non-navigable areas, creating paths, developing autonomous navigation focused on mobile robotics, integrating georeferencing techniques, data processing images and navigation using reinforcement learning without the need for prior mapping of the environment. For this, software was developed that, through the current location of the robot using (Global Positioning System) GPS coordinates, captures satellite images of the region in which it is located, improving the characteristics of the image. captured of the region, thus identifying its main characteristics and detecting regions that may or may not be used for navigation, generating a map of the environment at the end of the process, and determining a possible path between the starting point and destination through the algorithm Rapidly Exploring Random Tree (RRT), determining the points on the map to be covered. These points are sent to the agent in which we have the displacement for each of these points established, working in two different ways. In the first case, when there is no obstacle, the robot follows the shortest distance between the starting point and the end, however, if there is an obstacle, the reinforcement learning algorithm (Q-learning) is used. Experiments were carried out to qualitatively and quantitatively evaluate the technique and, through the simulation results, the technique proved to be in line with the intended objectives. The main highlight of this work is the integration of algorithms so that we do not need to previously map the external environment with traditional techniques such as Simultaneous Localization And Mapping (SLAM), which would require more time, as it is necessary to go to the environment beforehand and scroll through it to later navigate. With the techniques developed in this work, any external terrestrial environment can be mapped and used for the development of various applications aimed at mobile robotics, serving as a basis and direction for practical implementations in future work.