— This paper addresses the multiple tasks performing issues for redundant nonholonomic mobile manipulators. An extended gradient projection redundancy resolution scheme is proposed which can determine the directions of self-motion to perform multiple secondary tasks. This scheme is easy to use and can avoid algorithm singularities. A general dynamic modeling method is presented in consideration of nonholonomic constraints, interactive motions and self-motions. A real-time fuzzy logic self-motion planner is devised to create desired selfmotion magnitudes and a robust adaptive neural-network controller is designed to accomplish multiple secondary tasks without affecting the primary one in the workspace. The effectiveness of the proposed algorithm is verified through simulations for a 3DOF manipulator atop a 3-wheeled mobile platform system.