Accurate large-displacement magnetic levitation actuation and its stability remain difficult in non-liquid environments. A magnetic levitation actuation and motion control system with active levitation mode is proposed in this paper. The actuating force of the system is generated by the external magnetic field. A neural network proportion-integration-differentiation (PID) controller is designed for active actuation, and a force imbalance principle is built for the step motion mode. Dual electromagnetic actuators are configured to generate a superimposed magnetic field, ensuring that the electromagnetic force on the ball is more uniform and stable than single actuators. Dual-hall-structure sensors are used to measure displacement, thereby reducing overshoot and ensuring stability whilst motivating the ball. Due to the high adaptability of the neural network to complex systems with nonlinear and ambiguous models, the PID controller composed of neurons has stronger adaptability through tuning the PID controller parameters automatically. Furthermore, the proposed controller can solve the shortcoming that the deviation between the controlled object and the steady-state operating point increases and the tracking performance deteriorates rapidly. The strong robustness and stability in active levitation and motion control is achieved during both ascending and descending processes.