The movement control of the cerebellum and its adaptability to the environment are the keys to complete rapid and precise movement for humans. Simulating and studying the operating mechanism of the cerebellum will provide a better way to control complex and changeable robot models. Therefore, by following the real biological ratio of different types of cerebellar neurons, a large-scale spiking neural network model of the cerebellum is built and the realistic structure, information transmission method, and learning mechanism of the cerebellum are simulated. We also complete the error correction control of a simulated robotic arm, and clarify the influence of different synaptic plasticities on the control effect of the cerebellar network with the control results of the system under different control tasks. In order to further increase the biological authenticity of the cerebellum control system, the model is implemented on a field programmable gate array (FPGA) platform in a parallel operation approach closer to the human brain, and corresponding resource optimization methods are proposed so that the achievable network scale is increased. The FPGA implementation results show that the system successfully simulates the adaptive brain-inspired robotic arm control based on the cerebellar error correction ability. The cell dynamics of the cerebellum can also be reproduced on the system and the high fault tolerance from large-scale granule cells is proven. This work provides a platform that takes into account both the realization of cerebellar application functions and the theoretical research.