Title: Position Control of a Quadrotor via Soft Actor-Critic Reinforcement Learning for Thrust Vectoring

Abstract: This study introduces an innovative Reinforcement Learning (RL) framework for quadrotor control. While existing research predominantly targets the direct regulation of individual rotor RPMs, this work focuses on managing the quadrotor’s thrust vector. The RL agent is tasked with determining the proportional thrust contribution along the vehicle’s z-axis, as well as the target Roll ($\phi$) and Pitch ($\theta$) angles. These computed control commands, combined with the current Yaw angle ($\psi$), are transmitted to an attitude PID controller, which subsequently translates them into specific motor RPMs. The Soft Actor-Critic (SAC) algorithm—a model-free, off-policy, stochastic RL method—was employed to train the agents. Experimental data indicates that the proposed thrust vector controller achieves significantly shorter training durations compared to traditional RPM-based controllers. Furthermore, simulations demonstrate that this approach yields superior path-following performance, characterized by enhanced smoothness and precision.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC