The study aims to improve the vibration isolation effect of the semi-active seat suspension system using a reinforcement learning control scheme. Mechanical tests on the designed magnetorheological damper to establish the mechanical model of the damper. According to the parameters of the seat suspension, the modeling of the magnetorheological semi-active seat suspension system is derived, and a control scheme based on reinforcement learning is proposed. Finally, the performance of the proposed control strategy is compared with the passive system and the sky-hook control under random single. It is shown that the RMS value of acceleration of the seat suspension is reduced by 21.2% compared to the sky-hook control, and by 10.6% compared with the passive system.