Towards monocular vision-based autonomous flight through deep reinforcement learning

作者：

Highlights：

• A reinforcement learning-based obstacle avoidance algorithm for small drones.

• A monocular vision-based obstacle avoidance using the estimated depth information.

• A newly-designed reward function for the fast and safe algorithm.

• Applying the learned policy from simulations to real worlds without further training.

摘要

•A reinforcement learning-based obstacle avoidance algorithm for small drones.•A monocular vision-based obstacle avoidance using the estimated depth information.•A newly-designed reward function for the fast and safe algorithm.•Applying the learned policy from simulations to real worlds without further training.

论文关键词：Obstacle avoidance,Depth estimation,Vision-based,Deep reinforcement learning,Q-learning,Navigation decision making

论文评审过程：Received 27 April 2021, Revised 10 November 2021, Accepted 22 February 2022, Available online 9 March 2022, Version of Record 15 March 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.116742