The railway industry recently showed interest in the potential use of AI to render trains autonomous in order to reduce cost and improve security and performance. This paper focuses on the integration of AI into Automatic Train Operation (ATO) systems to control train speed. The objective of this paper is to present and analyze a review of the literature made in that context. The review is done according to a typology based on three axis: the inputs and objectives of the model, the AI method used by authors and last, the validation process. Our review shows that AI based approaches outperform classical approaches and that learning based methods are superior to rule-based systems. Meanwhile, the contributions present incomplete validation processes, difficulties to generalize the proposed AI method and last, a lack of use of perceptual data during decision making. This analysis enables us to draw some prospects relevant to the solving of the listed limitations.