The adaptive biasing force (ABF) scheme is a powerful molecular-dynamics based method for overcoming barriers of the free-energy landscape. Integration of the mean force measured along a chosen reaction coordinate (RC) yields the so-called potential of mean force (PMF). The RC is a coarse-grained description of the transition mechanism. The mean force is estimated by accruing and averaging the instantaneous force exerted on the system. The PMF is then used to bias the standard dynamics of the system in order to improve sampling in the RC. We show that faster exploration of the reaction pathway can be achieved by running multiple walkers in parallel and exchanging information at fixed intervals in the course of the simulation. Numerical experiments performed on the prototypical deca-alanine peptide demonstrate that the convergence properties of the free-energy calculation are globally improved through a more efficient exploration of compact configurations reflected in parallel valleys of the free-energy landscape. Diffusion along the RC is further enhanced by a selection mechanism, whereby far-reaching walkers are cloned, replacing less effective ones.