Recently, the potential for assistive devices to enhance the walking activity of the elderly has increased. Elderly have muscle weakness in their lower limbs and they are liable to persistent injuries because of falling on the ground. Lately, different mobility aid robots, especially canes, have been proposed to improve their walking activity. The complicated design and the unsuitability for real-time fall prevention are two main demerits of the proposed cane robots in literature. In this paper, it is proposed to develop a simple yet effective 4-DOF cane robot for real-time fall prevention of elderly to work in two modes namely, normal walking and fall prevention. It consists of a stick that is connected to an omni-directional mobile platform by a simple revolute joint. In the normal walking mode, the cane robot is kinematically controlled to follow the human in the front and achieve a desired distance and orientation apart from the human. The quite short response time that is required to prevent the elderly falling by the cane robot is a challenging problem. So, based on the full dynamic model in the 3D space, an impedance controller whose parameters tuned by the genetic algorithm with force/torque constraints is developed. Additionally, for the sake of using the proposed cane robot by different users with different heights and interactive forces, a model reference adaptive control is imposed to the cane robot to overcome the system uncertainties. The results prove that the four DOFs are enough to prevent falling and the kinematic controller achieves the follow-in-the-front concept accurately as well as the dynamic responses are suitable for real-time fall prevention and ensure the robustness against the uncertainties.