This paper presents a methodology to design an optimized Lane Centering assistance (LCA) system for an imposed structure. It proceeds within the H 2 /H ∞ framework to manage the trade-off between lane centering and comfort. At a first stage, convenient indicators are defined in a way to be understandable, closely linked to practical specifications and easy to compute. Some of them require using exogenous signal model (e.g. curves or gusts of wind), giving a priori information on possible scenarios. Based on them, a multi-scenarios and multimodel robust control problem is built, closely linked to the initial requirements. It aims to minimize the jerks of the lateral error and the steering wheel induced by the disturbances, under constraints of performance and robustness specifications. Deterministic non-smooth optimization algorithms are considered to solve it. The efficiency and the robustness of the synthesized controller is demonstrated by simulation using real measurements of a camera. Lastly, the proposed methodology makes future adaptation to other ADAS easier, which is a crucial point in an industrial context.