The control logic of a reconfigurable meta-surface in mmWave is investigated in order to perform real-time beam steering in mobility contexts. When it is necessary to track an object or a person in movement, it is required to change the direction of the signal transmitted/reflected by the metasurface. To do so, the meta-surface has to be reconfigured to modify the generated radiation pattern. Here a specific metasurface, working around 78.5 GHz, is adopted, which consists of many unit-cells containing a diode to make its electromagnetic behaviour reconfigurable. In order to enable the system to adaptively compute the new coding schemes, namely to change the states of the meta-surface unit-cells, a machine-learning algorithm is adopted. In particular, we propose a Genetic Algorithm, integrate it in the architecture of our meta-surface and evaluate the performance. Results show that a good accuracy and a good convergence time are achieved for real-time beam steering functionality and the capacity of the system equipped with RIM (reconfigurable intelligent metasurface) is increased.