In this paper, we investigate human generated bang-bang inputs that can switch equilibriums of nonlinear dynamical systems. For this purpose, we propose an experimental system in which the human can generate the bang-bang input to switch the four equilibriums of a coupled inverted pendulum model. We numerically obtain reachable sets of given equilibriums to characterize the switching time of the bang-bang input to reach the equilibrium. Then, we perform an experiment in which human subjects manually optimize the bang-bang control inputs. The result shows that the motion strategies generated by the subjects become improved as the number of trial increases and that such optimization is not unique and can be classified into certain categories.