Exploring Urban UAV Navigation: SAC-Based Static Obstacle Avoidance in Height-Restricted Areas Using a Forward Camera

The proliferation of Unmanned Aerial Vehicles (UAVs) in urban settings presents significant navigational challenges, particularly in obstacle-rich environments. When UAVs operate in height-restricted urban areas, obstacle avoidance navigation becomes unavoidable. This research investigates the efficacy of employing only a forwardfacing camera in conjunction with the Soft Actor Critic (SAC) algorithm for static obstacle avoidance in height-restricted areas. This minimalist sensor approach addresses cost and weight constraints while ensuring navigational safety. Our UAV was configured to navigate through a simulated urban landscape, created using the Gazebo simulation platform, which mimics real-world urban height-restricted complexities such as underground parking lots. Through extensive training and evaluation, the SAC enhanced system demonstrated robust obstacle detection and avoidance capabilities. Results indicated our DRL-based obstacle-avoidence method is effecient under urban environment even with limited sensory input.