Modeling of Unsafe Areas for Swarm Autonomous Agents

Abstract

In the motion tasks of an autonomous vehicle it is necessary to consider the position of individual obstacles and various unsafe zones. We consider a multi-agent system whose motion is carried out towards a common goal according to algorithms of swarm behavior based on Reynolds rules: speed matching, collision avoidance with neighbors and attraction to neighbors. Three approaches to modeling swarming behavior based on articles (Olfati-Saber, Flocking for multi-agent dynamic systems: algorithms and theory. IEEE Trans Autom Control, 51(3), 2016; Olfati-Saber, Murray, Flocking with obstacle avoidance: cooperation with limited communication in mobile networks. In: 42nd IEEE international conference on decision and control, vol 2, 2022–2028) are presented. The peculiarities of this work are that the approaches considered together help to realize such a model of motion, which ensures the avoidance of collision with all available types of obstacles. The method is intended for implementation in those spaces, where there will be many autonomous vehicles. The main problem is that when a dynamic obstacle is encountered, congestion can occur—the agents closest to the obstacle react quickly, while distant agents do so with a lag and create crush. The goals of this paper are to propose a method of indirectly transmitting danger information between swarm members without using communication channels. This means that those robots that do not see the danger can get information about it from other agents by observing their behavior. For this purpose, a method of escaping from a pack from a “predator” based on Q-learning is implemented.