Semi-Active Reinforcement Learning Suspension Control for the Off-Road Vehicles

Vertical vibration of the terrain vehicle involves its comfort, maneuverability, and fatigue life of the key components. The type of road surface encountered by the vehicle is very complex and difficult to measure. Besides, the damper system has a strong non-linearity, which makes the design of the vehicle suspension controller difficult. Compared with the active suspension, the semiactive suspension has the advantages of low energy consumption and high safety, therefore this paper uses semi-active magneto-rheological damper and reinforcement learning technology, from the comfort point of view, to solve the suspension random optimal control problem. It is expected to improve the comfort of the terrain-vehicle with intelligent, low-power control method. This paper applies Gaussian Process (GP) technology to learn and model the nonlinear part of the terrain vehicle system, and then carries out the design of the reinforcement learning control strategy, establishes a linear control law with the suspension deflection, the sprung mass velocity, and the unsprung mass velocity as the feedback variables. The introduced reinforcement learning algorithm learns the appropriate feedback gain during the interaction process with the system, and then uses the learned feedback gain to carry out system simulation and experimental analysis. The paper compares the reinforcement learning algorithm with other classical semi-active control algorithms under the input of random, bump and sinusoidal pavement, and the analysis results show that the reinforcement learning algorithm introduced in the paper has excellent control effect in the full frequency band and can provide good comfort for terrain vehicles under the condition of low power consumption.