Genetic Policy Search involves performing Local Policy Search, but starting from a plurality of initial policies and perturbing the top-k most successful ones (called the “elite samples” m_{elite}) to generate the next set of starting points.