An executive system for cognitive agents

We present the executive component of the GNOSYS cognitive agent architecture that is responsible for regulating the behaviour of a robotic agent. The system handles a number of problems such as: the execution of multiple goals, dynamic priorities of goals, opportunistic execution, elimination of non-converging goals and other issues. We provide the basic ideas behind the current design and explain the various mechanisms that allow a complex but flexible behaviour for a robotic agent. Simulation results provide not only evidence for efficiency of the design but also help clarify the operation in a variety of problems. Experiments with a real robot in the laboratory are also presented. We conclude with a brief comparison to other existing systems and comments on future work.