Enhancing skyhook for semi-active suspension control via machine learning

Semi-active control is the most employed technology for electronic suspension systems. The damping can be regulated to trade-off comfort and handling. Due to its success in industrial applications, semi-active control design has been extensively investigated in literature mainly from a model-based perspective. In this contribution, the authors propose a novel control strategy derived via a sequential learning framework, which selects the most significant feedback measurements for semi-active control and learns the optimal policy from data. As opposed to most of the contributions based on deep-learning approaches, the output of the proposed methodology is a control algorithm consisting of few parameters, which can be easily ported and calibrated on a real vehicle. Experimental validation on a sports-car shows that the proposed algorithm is superior in damping the body resonance with respect to the state-of-the-art skyhook algorithm. Indeed, the learned control policy consists of an augmentation of skyhook.