PPO Reinforcement learning for smooth control of two-wheeled mobile

This paper proposes a reinforcement learning (RL) approach to improve trajectory tracking in two-wheeled mobile robots, which are difficult to control due to nonlinear dynamics and nonholonomic constraints. Unlike traditional methods such as sliding mode control, the proposed strategy uses the proximal policy optimization (PPO) algorithm to map robot state position, orientation, and tracking error directly to velocity commands. The reward function encourages accuracy, smooth motion, and energy efficiency. Simulation results show that the RL controller matches the accuracy of a baseline sliding mode controller (SMC) while producing smoother inputs and avoiding chattering. It also generalizes well across various trajectories without retuning. This demonstrates RL as a robust, adaptive alternative to model-dependent methods, with future work aimed at hardware testing and hybrid RL-classical control designs.