APPROACH TO BUILDING A GLOBAL MOBILE AGENT WAY BASED ON Q-LEARNING

Abstract

Today, the problem of navigation of autonomous mobile systems in a space where disturbances are possible is urgent. The task of finding a route for a mobile robot is a complex and non-trivial task. At the moment, there are many algorithms that allow you to solve such problems in accordance with the specified criteria for building a route. Most of these algorithms are modifications of "basic" path planning methods that are optimized for specific conditions. The subject of research in the article is the process of building a global path for a mobile agent. The purpose of the work is to create an algorithm for planning the route of autonomous mobile systems in space using the Q-learning algorithm. The following tasks are solved in the article: development of an approach to training and support of a reinforcement learning algorithm for building a global path of a mobile agent; testing the agent's ability to find a path in environments that are not in the training set. The following methods are used: graph theory, queuing theory, Markov decision-making process theory and mathematical programming methods. The research is based on scientific articles and other materials from foreign conferences and archives in the field of machine learning, deep learning and deep reinforcement learning. The following results were obtained: an approach was formulated to construct the global path of a mobile agent based on the accumulated data in the process of interaction with the external environment. The environment rewards these actions and the agent continues to carry them out. This approach will allow this method to be applied to a wide range of situations and devices. Conclusions: This approach allows accumulating the knowledge of the outside world for further decision-making when planning a route where the robot can acquire the skill of self-learning, studying and training like a human, and finding the path from the initial state to the target state in an unknown environment. In the modern world, the use of robots and autonomous systems is spreading, designed to replace or facilitate human labor, make it safer and speed it up. Adaptive autonomous path finding algorithms are very important in many robotics applications. Thus, navigation tasks with limited information are relevant today, since this is the main task that the agent solves, and one of the tasks that are part of the robot during operation.

Keywords

• path planning
• Q-learning
• mobile works
• adaptive standalone search algorithms