Reinforcement Learning-Based Simulation of Semi-Active Suspension Systems in Quarter-Car Model

Abstract

This study investigates the application of reinforcement learning (RL) techniques for simulating and managing semi-active suspension systems within a quarter-car model framework. Conventional controllers, such as PID and skyhook, typically depend on fixed parameters and simplified system assumptions, which can restrict their effectiveness when faced with nonlinear and variable road conditions. RL, on the other hand, provides a flexible, model-free control methodology that learns optimal strategies through continuous interaction with the vehicle’s dynamic environment. In this work, three RL algorithms - Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Deep Deterministic Policy Gradient (DDPG) are utilized to reduce body acceleration and tire displacement, thereby enhancing both ride quality and handling performance. Simulation studies conducted on a range of road disturbances, including sinusoidal, step, and random profiles, reveal that RL-driven controllers outperform traditional approaches in terms of adaptability, robustness, and smooth control actions. The results highlight RL’s capability to deliver efficient suspension control without requiring explicit system models or extensive manual tuning, demonstrating strong generalization to previously unseen conditions. These benefits underscore the potential of reinforcement learning as a powerful tool for developing intelligent, autonomous suspension systems in modern vehicles.