Depth-bounded logic for realistic agents