This paper describes an attendant service robot using RT Middleware. We use laser range finders(LRFs) for human detection and tracking. Legs of people are first detected in LRF data and then temporally integrated with a linear motion model in order to estimate and predict the people positions. For each person, we compare his/her predicted position with detected leg positions, and choose the one with the shortest distance as the person's leg and update the position. We set three target positions as both sides and the back of the predicted target person position, and the robot usually uses one of the side positions and switches to the back position if it is necessary to avoid collisions, thereby realizing a safe following behavior. We developed a user interface that is composed of a web browser-based program and a server program running on the robot, by which the user can control the robot manually or switch the control modes with a smartphone.