Simulation-based comparison of robust control and deep reinforcement learning control techniques for active suspension systems

Modern vehicular systems rely heavily on suspension performance for ride comfort, handling stability, and safety. Conventional passive suspensions, while common, cannot adapt to changing road conditions. Active suspension systems overcome this limitation by using actuators and control algorithms to adjust suspension behavior in real time based on sensor feedback, making them especially valuable for vehicles operating in unpredictable environments. This paper explores the design and comparative evaluation of various active suspension control strategies, including H−∞ synthesis, μ synthesis, and reinforcement learning-based methods. A quarter-car model with real-world parameters was used for simulation, implemented in both MATLAB and Python. The controllers were tested under multiple excitation profiles: a single hump, five consecutive humps, and standardized ISO 8608 road profiles that emulate realistic terrain variability such as random bumps and potholes. Each control strategy was assessed using performance metrics such as body acceleration and suspension deflection. The results demonstrate that all active suspension systems outperform passive suspensions, with the reinforcement learning controller providing the greatest overall reduction in body displacement.