A Comparative Study of LQT Controller Design Using Reinforcement Learning Methods: A Case Study on Speed Control of the PMDC Motor

This paper presents a comparative analysis of Linear Quadratic Tracking (LQT) controller design using various Reinforcement Learning (RL) methods, applied specifically to speed control in Permanent Magnet Direct Current (PMDC) motors. Traditional LQT controllers typically depend on offline design and system modeling, limiting their adaptability in real-time applications. To overcome these limitations, we investigate three RL methods and evaluate their performance in terms of convergence rate, computation time, reliance on system models, and steady-state error. Furthermore, the study compares RL-based LQT controllers with conventional PID controllers. Results show that RL methods, particularly Q-learning, provide a model-free approach to optimal control design, achieving zero steady-state error and no overshoot, offering significant improvements over traditional PID methods.