A Robust and Intelligent Obstacle Avoidance Method for Uncrewed Aerial Vehicles Under the Complex and Dynamic Environment

Unmanned aerial vehicles (UAVs) are of great importance in the confined spaces for executing diverse tasks. However, UAV control in those environments remains heavily reliant on human, since it needs to avoid obstacles under the complex scenarios where UAVs are subject to both static objects and dynamic pedestrians. Current dynamic obstacle avoidance schemes cannot be well adapted to these environments. To address these challenges, we propose a robust and intelligent obstacle avoidance method based on active perception of environment uncertainty. An environment-perception network is designed to enable the drone to perceive and encode the dynamic environment, and reinforcement learning is utilized to facilitate effective avoidance of dynamic pedestrians in the presence of observational uncertainty. Simulation experiments demonstrate that our method outperforms existing computational and deep learning approaches in dynamic scenarios, surpassing human testers.