This work presents a method to generate optimal trajectories for redundant mobile manipulators based on a weighted function that considers simultaneously joint torques, manipulability and preferred joint angle references. This method is applicable to a group of tasks, commonly known as push-pull tasks, in which a redundant mobile manipulator subject to non-holonomic constraints moves slowly while exerting a set of forces against the environment. In practice, this occurs when the manipulator is pulling against an object such as when opening a door or unearthing a buried object. Torque is computed in a quasi-static manner, mainly taking into consideration the effect of multiple external forces while neglecting dynamic effects. The formulation incorporates a criterion for optimizing a starting configuration, and special considerations are made to account for non-holonomic constraints. The application to an existing mobile manipulator is described.