Abstract—This research implements to optimize gait behavior of a biped robot while walking on rough road. Gait pattern is produced by solving constrained optimization problem using improve self-adaptive differential evolution (ISADE) and response surface model (RSM). The foot structure of the robot adopts a toe mechanism and is generated by topology optimization in which four specific situations are considered. The optimal structure is a combination of these four results. This structure not only reduces the weight of the robot but also ensures its stable walk. In addition, while considering the human walk, we discover that arm swing motion can preclude ground reaction torque caused by leg wing, thus, we applied a mechanism to imitate arm swing motion of the humans for the robot to enhance stability. It can be said that with two applied mechanisms, the robot motion is primarily comparable to the human one. Our result is validated by dynamic simulation in Adams environment (MSC software).