Optimizing quadrotor navigation through emotional deep reinforcement learning: leveraging emotional rewards and states for enhanced training efficiency

Abstract

Deep reinforcement learning (DRL) has revolutionized artificial intelligence by enabling agents to learn optimal behaviors through interactions with their environment, and has been applied to complex systems such as quadrotor navigation. However, conventional DRL primarily emphasizes logical intelligence and often overlooks emotional factors that influence human decision-making. This study introduces emotional deep reinforcement learning (EDRL), a novel framework that integrates emotional rewards and states specifically anger and pleasure into the learning process. Drawing inspiration from human emotion-driven decision-making, EDRL improves training efficiency, reducing the number of episodes that needed to achieve optimal performance. These findings highlight the critical role of emotional factors in autonomous learning, offering new insights into the development of more efficient, human-like, and emotionally intelligent AI systems for quadrotor navigation in complex and uncertain environments. © The Author(s), under exclusive licence to The Brazilian Society of Mechanical Sciences and Engineering 2025.