This paper propose a new control algorithm for a Robotic Walking Support System, which varies the kinematic structure of the system to adapt users' walking or controlling disability. Elderly who suffers from deficiencies in motor skills cannot control or steer their walkers properly. Since the applied torque is highly correlated to the heading angle of the walker, torque transformation is needed to properly steer the walker. The variation of kinematic structure will be implemented by varying the Center of Rotation (COR) and as a result the torque of the system is changed. In order to model the users' walking or controlling disability, the user will be asked to follow a desired path. Using a learning algorithm, errors between the desired and actual path will be used to determine the new COR. The proposed control algorithm was experimentally implemented and the result shows its validity.