Motion-force control of redundant manipulators is essential in realistic applications, attracting significant research attention in recent years. However, there are two major issues in existing works. Firstly, current kinematic-based motion-force control schemes primarily focus on positional control, neglecting the control of force’s orientation. Secondly, these methods are not suitable for redundant manipulators with inaccurate kinematic parameters, as the inaccurate Jacobian matrix deteriorates the control performance. In response to these challenges, this paper proposes a motion-force control scheme for redundant manipulators with inaccurate kinematic parameters, addressing the learning and motion-force control from both position and orientation perspectives. Initially, a unified motion-force control method is developed from a kinematic perspective. Subsequently, an axis-angle representation-based orientation control method is introduced. Additionally, learning formulas are devised to estimate the Jacobian matrices required for motion-force control and orientation control. Ultimately, the learning and control processes are unified into a quadratic programming problem solved by a neural dynamic controller. Theoretical analyses, simulations, and physical experiments validate the effectiveness and novelty of the proposed scheme and controller.