In the application scenarios of quadrotors, it is expected that only part of the obstacles can be identified and located in advance. In order to make quadrotors fly safely in this situation, we present a deep reinforcement learning-based framework to realize autonomous navigation in semi-known environments. Specifically, the proposed framework utilizes the dueling double deep recurrent Q-learning, which can implement global path planning with the obstacle map as input. Moreover, the proposed framework combined with contrastive learning-based feature extraction can conduct real-time autonomous obstacle avoidance with monocular vision effectively. The experimental results demonstrate that our framework exhibits remarkable performance for both global path planning and autonomous obstacle avoidance.

Stratospheric airships have significant potential for remote sensing and communication, but their complex spatiotemporal environment poses significant path planning challenges. Our approach models the environment as the Cartesian product of diverse meteorological field time series and airship states, establishing a framework that accommodates more complex constraints in path planning. Additionally, by considering the impact of cold cloud fields on airship pressure differentials, our method enhances structural stability and safety by guiding the airship away from cold cloud regions. The method integrates the Soft Actor-Critic algorithm with the Long Short-Term Memory network, refining the model's exploration capabilities and its ability to capture temporal dependencies in sequential data. Comparative analyses indicate that the proposed model consistently outperforms alternative models across all evaluated metrics.