This paper presents a foot structure for humanoid robot locomotion on rough ground called an adaptive foot which is a novel structure proposed in our previous study when considering the robot locomotion on flat ground. Realizing stable walking on rough ground is essential, because humanoid robots are expected to assist humans in not only a flat known domestic environment but also on rough ground surfaces with terrain variances while maintaining their tasks. The mentioned foot structure equips the robot with a good adaption. It enables the foot to increase the contact points and improve the stability on complex surface. Moreover, this structure also reduces the effect of impact force on the robot foot at toe off period in the walking. In this study, a gait pattern is generated by an approximated optimization method based on Response Surface Model (RSM) and Improved Self-Adaptive Differential Evolution Algorithm (ISADE). The result is experimentally validated through dynamic simulation on Adams (MSC software, USA) with the Kondo robot.