Toward complete coverage planning using deep reinforcement learning by trapezoid-based transformable robot