In this paper, the result of applying a distributed approach to large scale block stacking problems is shown. In the approach, each block is a mobile autonomous agent and solutions of the problems are obtained as a result of local interactions among the agents. The interactions are mediated by motivation values which evaluate agents' mobility with respect to individual goal attainment. Each agent has a built-in ANN behavior function which is trained by an EP technique. The training is carried out using small size problems and then, by applying the result, large scale problems up to one thousand blocks are solved.