Learning-based Control for Quadrotor Vertical and Slanted Landing Operations under Ground Effect Disturbances

In this study, we present a learning-based control framework for quadrotor vertical and slanted landing operations under ground effect disturbances. We develop and evaluate two reinforcement learning (RL) approaches that jointly learn to regulate thrust and compensate for aerodynamic disturbances induced by ground effect. The first approach employs a Deep Deterministic Policy Gradient (DDPG) agent trained to generate thrust commands for vertical takeoff and landing. This is accomplished by using state information to implicitly account for the proximity to the ground. The second approach builds on this framework using an invariant RL formulation, enabling policy generalization across slanted approach geometries that introduce asymmetric and partial ground effect interactions. In both cases, the RL agent operates as an end-to-end feedback controller that outputs thrust directly, without reliance on explicit aerodynamic models or disturbance observers. Simulation results indicate that reinforcement learning enables effective adaptation to ground effect phenomena and supports stable maneuvering in proximity to complex terrain during landing and takeoff.