This thesis explores a data-based approach to the problem of Friction Potential Estimation of Tyre-Ground systems. A Neural Network is specifically tailored to reliably draw conclusions on the available Friction Potential. A performance baseline is extrapolated from the evaluation of an LSTM Neural Network applied to the same problem. The architecture, chosen as a starting point for this job is, a Mono Dimensional Convolutional Neural Network. From there on, structural enhancement, such as multiple output training, are added to the mix. The dataset has been built by measurements gathered from a real-life vehicle, which has been run on a Test Bench.