Deep RL Quadrotor Control for Under-Canopy Forest Inspection

A new study from arXiv presents a deep reinforcement learning (RL)-based quadrotor controller designed for autonomous aerial inspection in under-canopy forest environments. The end-to-end control policy maps states directly to rotor RPMs, enabling simultaneous position and yaw reference tracking for inspection view-pose behaviors and point-to-point navigation. To ensure safe long-range deployment, the system integrates a higher navigation layer comprising a Traveling Salesman Problem (TSP) planner for optimal visitation sequencing and a Rapidly-exploring Random Tree Star (RRT*) planner for path planning between target regions. The approach addresses the challenge of operating in cluttered, GPS-denied forest environments where traditional controllers may fail. The paper is available on arXiv under reference 2605.19202.