The gecko increases its speed by using the waist joint. We thought this mechanism might be effective for consistent stability and speed of a robot, and we designed and prototyped gecko-type four-legged robot to verify this. In this study, a walking acquisition experiment that uses reinforcement learning was conducted to verify the best operation in this mechanism, and the influence that the waist joint had on the walking activity was examined. Moreover, because large right- and left-gaps in the walking orbit were caused by having the waist-joint mechanism, we conducted orbit correction by changing the reward function in reinforcement learning.