AUTONOMOUS SEARCH OF A TARGET WITH UNKNOWN COORDINATES AND SIMULTANEOUS VOXEL MAPPING OF THREE-DIMENSIONAL LABYRINTH-LIKE ENVIRONMENTS BY A SWARM OF UNMANNED AGENTS

Abstract

This paper addresses the problem of autonomous target search with unknown coordinates and simultaneous voxel occupancy grid mapping of complex 3D labyrinth-like environments by a swarm of unmanned agents in GPS-denied scenarios. A fully decentralized, reactive swarm architecture is proposed, which builds upon classical Reynolds flocking rules, artificial potential fields, and a permanent collective spatial memory mechanism that drives the agents toward the exploration frontier. To mitigate swarm stagnation in deep recursive pockets and narrow passages, an escape metaheuristic called "Amoeba" is introduced, enabling kinematic wall-following behavior upon detecting local stagnation. Simulation experiments across various landscapes evaluate space coverage dynamics relative to swarm size and environment tortuosity. The results confirm an asymptotic rather than linear decrease in exploration time due to the overlap effect, alongside a sharp non-linear increase during the final percentages of mapping, known as the last-mile problem. Crucially, the activation of the "Amoeba" escape strategy is shown to prevent practical system failure and significantly accelerate mapping efficiency in topologically complex spaces.