This paper studies the manipulation of large-scale systems of active particles, specifically addressing challenges faced in diverse fields such as crowd management, robot swarm control, and material transport. To address the scalability and robustness issues of existing methodologies, we develop an effective leader-based control strategy that combines reinforcement learning (RL) and artificial forces. This study introduces the generalized Vicsek model to explain how a leader guides active particles and applies it to the large-scale evacuation problem using a robotic rescuer (leader). Our results demonstrate a robust and efficient evacuation strategy that overcomes the inefficiencies of RL alone.