Simulated Annealing-Deep Deterministic Policy Gradient Algorithm For Quadrotor Attitude Control

Quadrotors have recently become one of the most popular and widely used types of unmanned aerial vehicles (UAVs) because of their distinctive applications. This category of vertical take-off and landing (VTOL) UAVs presents many properties, such as high maneuverability, compact size, simple propulsion system, and structural design. However, the under-actuated and nonlinear dynamics of these vehicles require the adoption of quite sophisticated control strategies for stability and trajectory tracking. This paper addresses the quadrotors attitude control problem using a Reinforcement Learning (RL) approach. To that end, a new Simulated Annealing-Deep Deterministic Policy Gradient (SA-DDPG) algorithm is being developed to achieve high control performance in a variety of scenarios. Furthermore, the proposed reward function is designed to speed-up the learning process. The obtained results demonstrate the efficiency of the proposed approach in both learning stability and attitude control competency.