This paper focuses on a learning observer-based fault-tolerant tracking control scheme for robot manipulator system with parameter uncertainty, actuator faults and input saturation. Without prior knowledge of faults, a learning observer is introduced for estimating the angular speed of a robot manipulator and reconstructing faults. Subsequently, through an introduced auxiliary function and fault estimation value, a fault-tolerant control law based back-stepping control was designed. The stability of the system was proved through the Lyapunov criterion, and the feasibility of the observer and controller was verified through numerical simulation.