A novel three-limbed robot was described and its motion planning method was discussed. After the introduction of the robot mechanical structure and the human-robot interface, a two-grade search mechanism based motion planning method was proposed. The first-grade search method using genetic algorithm tries to find an optimized target position and orientation of the three-limbed robot. The second-grade search method using virtual compliance tries to avoid the collision between the three-limbed robot and obstacles in a dynamic environment. Experiment shows the feasibility of the two-grade search mechanism and proves that the proposed motion planning method can be used to solve the motion planning problem of the redundant three-limbed robot without deficiencies of traditional genetic algorithm.