Deep reinforcement learning-based control of a 1-DOF helicopter: a comparative analysis of classical and modern methods

Abstract This study presents a comparative analysis of different control methodologies on the linearized model of a single degree-of-freedom (1-DOF) helicopter system around its operating point. The control structures evaluated in this study include a Proportional-Integral-Derivative (PID) controller optimized via Particle Swarm Optimization (PSO), Model Predictive Control (MPC), and the Deep Deterministic Policy Gradient (DDPG) algorithm based on Deep Reinforcement Learning (DRL). The conducted simulations revealed that the DRL-based controller exhibited a superior performance compared to the other methods, even on the linearized model, both in terms of reducing the magnitude of the error and improving the system’s transient response performance. This approach demonstrates the capacity to produce a rapid response without compromising the system’s stability, concurrently achieving faster rise and settling times. Furthermore, by effectively suppressing the time-weighted error values, it provides an advantage in terms of both accuracy and control quality. These compelling results indicate the significant potential of learning-based control methods in the control of nonlinear systems, establishing DRL-based strategies as a reliable foundation for advanced control applications and a superior alternative to classical methods.