Comparison of Reinforcement Learning and Evolutive Computing Learning techniques in drone swarm control

Reinforcement Learning and Evolutionary Computing techniques have proven to be powerful tools in solving a wide range of problems in fields such as robotics, logistics and path planning. In particular, they have been successfully used in trajectory optimization for unmanned aerial vehicles (UAVs). However, despite their potential, their application in drone swarm control has been remarkably scarce.

The use of drone swarms involves streamlining flight time and reducing operational costs, where, in addition to optimizing individual drone trajectories, it is crucial to ensure swarm cohesion, avoid collisions, and adapt to changing environmental conditions and mission demands. These challenges require innovative approaches that can learn and adapt in real time, which makes Reinforcement Learning and Evolutionary Computing techniques promising options.

Currently, most of the studies in the field of drone swarm control have focused on approaches with Reinforcement Learning, while the application of Evolutionary Computing algorithms is limited. This scarcity of cases can be attributed to the difficulty in designing a suitable representation of the drone swarm control problem that is compatible with these algorithms, or even to the fact that the focus in recent years has been mainly on the use of other types of algorithms related to Reinforcement Learning.

This study evaluated the effectiveness of Deep Q-Learning, Genetic Algorithm (GA) and Ant Colony Optimization (ACO) algorithms for route planning in drone swarms in 3x3, 5x5 and 7x7 dimensions. The results revealed that ACO significantly outperformed the other techniques in reducing actions and execution times, showing its value in contexts where operational efficiency is essential. Although Deep Q-Learning and AG also showed improvements, they failed to match the efficiency of ACO. These findings suggest that ACO is more suitable for practical implementation in complex missions, providing a solid basis for algorithm choice based on operational and task-specific demands. It is critical to highlight the importance of selecting the right algorithm and the optimal configuration of the number of drones to maximize the operational efficiency of swarms in real applications.