<p>Recent advancements in unmanned aerial vehicle (UAV) technology have enabled UAVs to be deployed as swarms to perform complex and critical tasks collaboratively. Bio-inspired algorithms play a crucial role in UAV swarms. It provides a distributed, decentralized method for exhibiting collective behavior. However, these algorithms are initially developed to exhibit simple, rule-based reactive and collective behavior. Applying these algorithms to develop a fully functional, decentralized multi-UAV system, such as a UAV swarm, presents significant challenges. These challenges include scalability, adaptability, rigidity, and robustness. This study presents a novel hybrid control framework named Q-learning enhanced Reynolds Flocking Algorithm (Q-RFA). The Q-RFA framework introduces a Q-learning meta-controller that selects among nine macro-strategies. The macro-strategies are defined using a Markov Decision Process (MDP) to dynamically adjust the flocking weights and enable adaptive swarming behavior. A Q-learning agent adjusts the flocking parameters based on a six-dimensional state of the swarm and environment. A complex, large 3D simulation is conducted to test the proposed framework. The Q-RFA is compared with the recent variant RFA algorithm to evaluate its effectiveness. The adaptive tuning approach significantly improves swarm robustness in cluttered 3D scenarios. Adaptive learning behavior enhances swarm stability, reduces collision rates, and improves responsiveness. While the framework is highly adaptive, it is also scalable. This makes it a promising strategy for real-world UAV swarm applications. However, it does not yet provide formal safety guarantees. It also does not account for real-world issues like communication delays or sensor noise. These limitations remain important areas for future work.</p>