Winter conditions, characterized by the presence of ice and snow on the ground, are more likely to lead to road accidents. This paper presents an experimental proof of concept of a collision avoidance algorithm for vehicles evolving in low adhesion conditions, implemented on a 1/5th scale car platform. In the proposed approach, a model-based estimator first processes the high-dimensional sensors data of the IMU, LIDAR and encoders to estimate physically relevant vehicle and ground conditions parameters such as the inertial velocity of the vehicle $v$, the friction coefficient $\mu$, the cohesion $c$ and the internal shear angle $\phi$. Then, a data-driven predictor is trained to predict the optimal maneuver to perform in the situation characterized by the estimated parameters. Experiments show that it is possible to 1) produce a real-time estimate of the relevant ground parameters, and 2) determine an optimal collision avoidance maneuver based on the estimated parameters.