Control Strategies for Two-Wheeled Balancing Robots: A Comparative Study of Robust, Nonlinear Control, and Reinforcement Learning

This paper presents a comparative study and analysis of classical and learning-based control strategies for a two-wheeled balancing robot, focusing on a Hierarchical Sliding Mode Controller (HSMC) and a Deep Deterministic Policy Gradient (DDPG) reinforcement learning policy. The primary objective is to enable stable balance as well as accurate velocity tracking in a dynamically unstable system subject to real-world disturbances. Both controllers were initially developed and validated in a simulation environment, where the robot was tasked with maintaining an upright position while tracking a reference forward velocity. The controllers were then deployed on a physical robot platform and experimental results demonstrated that while both controllers maintained overall stability, they exhibited some sensitivity to sensor noise and actuator latency. Nonetheless, the DDPG policy achieved effective tracking with minimal steady-state error, highlighting the potential of reinforcement learning for real-time control of underactuated robotic systems.