The cerebellar-model-articulation-controller Q-learning for the task assignment of a handling system
搬运系统作业分配问题的小脑模型关节控制器Q学习算法

The task assignment of a high-speed handling system with two robots is studied in this paper. In the underlying Markov decision process(MDP) model, the state variable is composed of both continuous and discrete values, and the state space is complex and suffers from the curse of dimensionality. Therefore, the traditional numerical optimization is prevented from successful application to this system. Since the cerebellar-model-articulation-controller(CMAC) has the advantages of fast convergence and desired adaptability, it is employed to approximate the Q-values in a CMAC-Q learning optimization algorithm for combining the concept of performance potential and Q-learning, and for unifying the average criteria with the discount criteria. Compared with the Q-learning, the proposed neuro-dynamic programming approach requires less memory, but provides higher learning speed and better optimization performance as shown in the simulations.