A framework for multi-agent cooperation applied to multiple robots

Cooperation between multiple robots is being used to tackle increasingly complex tasks. The amount of knowledge needed to build multi-robot systems is becoming unwieldy and it is difficult to engineer a closed system. One paradigm for overcoming this barrier is to organize the basic components of the systems in a manner that allows the sharing of performance, responsibilities, and resources. Until recently, research in multi-agent systems has been based on ad hoc models of action and interaction. The notion of intention is beginning to emerge as a prime candidate upon which a sound theory could be based. This research, motivated by questioning the suitability of the intentional theory, implements two of the most influential frameworks into a multi-robot domain to test their applicability and to identify any possible inadequacies. To refine the existing frameworks, this research proposes a new framework which is inspired by organisation theory and economic team theory. Shifting Matrix Management (SMM) divides agents' actions during the cooperation process into six stages, namely goal selection, act selection, team formation, plan formation and shifting. The mental states and the positions of agents in such a process are viewed as continually shifting in a matrix structure. To support the framework, a quantitative decision theory is developed. The theory reveals that an agent's mental state on selecting an action is defined by three quantitative functions: a probability density function, an outcome function, and a utility function. The framework has been formalised into a general and abstract model. The SMM model was implemented in the control of cooperation between multiple robots and a number of controlled experiments were performed to evaluate the model. The results indicate the effectiveness of the model. It is believed that the effectiveness of the SMM model may be generalisable.