Dynamic Control Comparison: Tuned-PID vs Trained Reinforcement Learning of the Double Inverted Pendulum on a Two-Wheeled Platform

This study offers a thorough and comprehensive examination of the double-inverted pendulum system, employing both PID control and reinforcement learning (RL) techniques. Through meticulous evaluation, the research compares the efficacy of these methodologies in stabilizing the system's dynamic behaviors. The Stand-up Routine Test meticulously assesses various performance metrics, including response times, overshoot values, and settling times, elucidating RL control's superiority in terms of agility and stability over PID control. Furthermore, the exploration extends into the Circular Routine Test, where RL control demonstrates the potential for achieving faster completion times, highlighting its adaptability in dynamic scenarios. Despite inherent variability, RL control consistently outperforms PID control, indicating its promising applications. The study underscores the dynamic nature of control methodologies, emphasizing the imperative for ongoing experimentation to optimize performance and practical implementation.