Autonomous quadrotor obstacle avoidance based on dueling double deep recurrent Q-learning with monocular vision

This paper proposes a novel learning-based framework to realize quadrotor autonomous obstacle avoidance with monocular vision. The framework adopts a two-stage architecture, consisting of a sensing module and a decision module. The sensing module trained in an unsupervised manner can extract depth information from the on-board camera image. Moreover, the decision module uses dueling double deep recurrent Q-learning to eliminate the adverse effects of the on-board monocular camera's limited observation capacity while choosing practical obstacle avoidance action. The framework has two advantages: (1) it enables the quadrotor to realize autonomous obstacle avoidance without any prior environment information or labeled datasets for training, and (2) its model can be easily updated while facing new application scenarios. The experiments in several different simulation scenes show that the trained framework outperforms a high passing rate in crowded environments and a good generalization ability for transformed scenarios.