A formal model for the coordination level of

intelligent machines is established. The framework of the

coordination level investigated consists of one dispatcher and

a number of coordinators. The model called coordination

structure has been used to describe analytically the

information structure and information flow for the

coordination activities in the coordination level. Specifically,

the coordination structure offers a formalism to (1) describe

the task translation of the dispatcher and coordinators; (2)

represent the individual process within the dispatcher and

coordinators; (3) specify the cooperation and connection

among the dispatcher and coordinators; (4) perform the

process analysis and evaluation; and (5) provide a control

and communication mechanism for the real-time monitor or

simulation of the coordination process. A simple procedure

for the task scheduling in the coordination structure is

presented. The task translation is achieved by a stochastic

learning algorithm. The learning process is measured with

entropy and its convergence is guaranteed. Finally, a case

study of the coordination structure with three coordinators

and one dispatcher for a simple intelligent manipulator

system illustrates the proposed model and the simulation of

the task processes performed on the model verifies the

soundness of the theory.