Free energy is a thermodynamic quantity driving important physical and chemical behavior in molecular systems. Insufficient sampling during finite length molecular dynamics (MD) simulations fails to capture all relevant configurations of the system making it impossible to accurately calculate absolute free energy using direct methods. An alternate approach is to perform restrained MD simulations at chosen centers in the collective variable (CV) space to calculate the mean force (negative gradient of free energy) and then recover the free energy surface from these values. It is important to minimize the number of points in CV space for mean force calculations to make the whole procedure computationally efficient. To achieve this goal, intermediate mean force surrogate model are constructed which drives the selection of future mean force centers. After the centers are chosen, the free energy surface is reconstructed using radial basis function approximation.