Abstract-Commercially available robots cannot always be adapted to arbitrary tasks or environments, particularly when the task would exceed the kinematic or dynamic limits of the robot. Modular robots offer a solution to this problem, since they can be reconfigured in various ways from a set of modules. The challenge of choosing the optimal composition for a given task, however, is hard since the search space of compositions is vast. Our approach addresses this problem: instead of finding the cost-optimal solution over all possible compositions individually, we propose a time-efficient composition synthesis method which uses evolutionary algorithms by taking taskrelated objectives into account. Simulations show that our algorithm finds the cost-optimal module composition with less computation time than other methods in the literature.