Study On Evaluation-Tradeoff Architecture

The existing action-selection architectures for an artificial creature have big insufficient in support for modularity and application targetability. They are also not good enough in robustness and transplant. In addition, they need to be improved in behaviors'coordination, behaviors'self-study, and behaviors'evolution. So we need to carry on deeper researches on action-selection architectures by establishing new theories, and adopting new methods. This paper analyzes the existing action-selection architectures, and sums up the foundational knowledge of researching in this field systematically. After introducting shema theory in cognitive psychology, discussing the relationship between schema theory and action-selection, and gathering up the relevant conclusions in neural science, which have the sense of guidelines to the researches on the action-selection architectures for an artificial creature, this paper puts forward evaluation-tradeoff architecture, a new action-selection architecture for an artificial creature based schema theory, and gives an exhaustive introduction from the following aspects: (1) Putting forward a new set of criteria which can guide architectures'design and implementation on the basis of generalizing the existing criteria for action-selection architectures for an artificial creature. (2) Abstracting behaviors into two grades: deliberation decision grade behaviors and exhibition performance grade behaviors, according to their different functions and different degree in intelligence and complexity on the basis of the conclusions in schema theory and neural science. (3) Being the first architecture that represents the behaviors in inner and external ways. Putting forward relation chart for behaviors'external representation and a new shema representation method for behaviors'inner representation. (4) Analyzing the shortcomings of the traditional potential field method, and putting forward a new behavior encoding method: dynamic potential field method. (5) Establishing evaluation-tradeoff mechanism for the deliberation layer's action-selection and action-cooperation, and establishing the reactive layer's action-selection and action-cooperation mechanism corresponding to the former based schema theory. Furthermore, generalizing the methods and processes for the design of an artificial creature's action-selection system based evaluation-tradeoff architecture. (6) Developping an simulation platform for autonomous football robots, and designing some football robots based evaluation-tradeoff architecture which performs well on the platform. At the end of this paper, a frame of knowledge-based evaluation-tradeoff architecture is put forward, which may be helpful to the successive researchers who are interested in reseaches on improved evaluation-tradeoff architecture in which behaviors'self-study, behaviors'self-adaptation and behaviors'self-evolution can achieve.