Robots are becoming increasingly accessible to both experts and non-experts. Therefore, establishing a method for learning robot operations that can be easily mastered by non-experts is important. With this in mind, we aimed to develop a method that facilitates skill acquisition for non-experts that operate robots. As a first step, this study examined the effects of button layout on the exploration and learning of robot operations. A humanoid robot was operated using an unfamiliar tablet-based user interface to achieve the task of shifting the robot's posture to the desired posture: single-foot-standing. The process in which participants found and repeated sequences of commands to achieve the shift task was observed. Four types of button layouts were prepared: normal, random, name appears after the first success (NAFS), and change to normal controller after the first success (CNFS). The normal layout roughly matched the position of the robot's joints, whereas the random layout was randomly assigned, and no information was displayed on each button. Before completing the shift task, a random layout was provided in the NAFS and CNFS layouts. After the first success, the corresponding joint information was displayed in the NAFS layout, whereas the layout was changed to a normal one in the CNFS layout. In total, 51 participants used the normal layout, 7 participants used the random layout, 25 participants used the NAFS layout, and 24 participants used the CNFS layout. The results indicate that providing a random layout during the exploration process (before the first success) is preferable for effective exploration and learning. However, during the learning process (after the first success), providing the relationship between joint movements and buttons in a visual manner is better without changing the button layout from that used in the exploration process.