Graph Grammars for Active Perception

The complexity of the applications in which robot sare being used does not stop growing. Different solutions such as sophisticated control architectures have been proposed in order to deal with complexity in robot control. These solutions make robotic systems more robust, scalable and easier to distribute,understand and monitor. However, it is still not clear how to cope with the complexity of the interaction dynamics that underlie the perception of the environment. With this issue in mind this paper presents the concept of cognitive graph grammarand two algorithms that make use of it. Cognitive graph grammars are a grammar-based theoretical framework designed to support cognitive perception and, especially, the active nature of perception. They provide a means to describe how graph-based models can be generated and the behaviors to execute depending on the perceptual context. This is done in such a way that the information provided using this formalism can be used for different perceptive purposes at the same time, such as to link action and perception or to diminish perceptive errors. The paper also describes an experiment in which a cognitive graph grammar is used in an autonomous robot in order to efficiently model an environment made of rectangular rooms with obstacles.