Aiming at the gait and balance disorder of the elderly with the increase of age and the occurrence of various senile diseases, this paper proposes a novel gait balance training robot (G-Balance) based on a 6-DOF parallel platform. Based on the platform movement and IMU wearable sensors, two training modes, active and passive, have been developed to achieve vestibular stimulation. Virtual reality technology is also combined to achieve visual stimulation. In the active training mode, the elderly actively exercise to control the posture change of the platform and the switching of the virtual scene. In the passive training mode, the platform movement is combined with the virtual scene to simulate some bumpy environments such as earthquakes to enhance the human anti-interference ability. In order to achieve smooth switching of the scene, continuous speed and acceleration of the platform motion are required in some scenarios, in which trajectory planning algorithm is applied to meet the requirement. This paper expounds the application of trajectory planning algorithm in balance training mode.