Enhancing Consumer UAV Autonomy: An Entropy-Augmented Deep Reinforcement Learning for Autonomous Navigation in Dense Obstacle Environments

Autonomous navigation for consumer unmanned aerial vehicle (UAV) faces growing complexity in cluttered environments like urban spaces. While deep reinforcement learning (DRL) has emerged as a key trend to enhance intelligent navigation capabilities in these settings, existing methods often struggle to balance exploration and collision avoidance amidst dense obstacles. To address this, we propose an entropy-augmented DRL framework integrating a state entropy-based intrinsic reward. This mechanism incentivizes UAV to explore under-visited regions, significantly boosting adaptability, decision-making robustness, and task efficiency. Extensive experiments across environments with varying obstacle densities demonstrate our method outperforms baseline DRL algorithms in both navigation success rate and task completion time.