We are willing to explore the possibilities to control several devices, such as mouse, keyboard, etc., which are around us, using robot arm. Among those devices, we apply machine learning theory in controlling the mouse. Mouse is a good example of devices which are hard to use control theories, because the relation ship between input (external force) and output (mouse pointer movement) is non-linear and difficult to define. For example, if we slide the mouse on the same side with the same force several times, the reactions of the cursor would be different according to the contact points. To experiment this subject, we used PUMA560 robot arm in the Robot Manipulation Lab. We can read the mouse cursor movement and get a lot of movement data for machine learning by letting the robot arm touch the mouse in several possible ways and directions. We applied two learning methods, which are linear regression and reinforcement algorithm to move the mouse to desired position. At section 2, we are going to introduce briefly about PUMA560 robot arm and our control interface. At section 3, we will compare linear regression and reinforcement algorithm in moving mouse cursor straightly. At section 4, we will use reinforcement algorithm in changing mouse cursor direction.