Copyright © IFAC TransportatioD Systems, TianjiD, PRC, 1994 A SYMBOLIC TRAFFIC CONTROL APPROACH A. BOULMAKOUL and S. SELLAM RV: Mathimatiques App/iquees et Intelligence Artificielle (M.A.I.A.) I.N.R.E.T.S., 2 avenue du General Malleret Joinville, 94114 Arcueil Cedex, France. Abstract. In this paper. we propose a way of formalizing the symbolic control of junctions. This approach is based on the one hand on the non-monotonic default logic and on the other hand on a co-operative multi-agent model. Our construction proceeds by analogy with the approach made by Reiter for logic diagnosis formalizing. In our way of formalizing. we propose a minimum action logic in order to make the control. Each agent of the junction (access controlled by a traffic signal) is described by a rational agent. The actors of the junction are led to co-operating in order better to achieve the control task. Key Words. Action logic. symbolic control. multi-agent system. non-monotonic logic. distributed intelligence. I. INTRODUCTION Control is an intelligent action situated in an anthropomorphic approach. Being intelligent leads to wanting to control one's environment (Foulloy. 1990). Different ideas of modelling system control have already been implemented; for instance the idea of automation with its deep (analytic) knowledge. the idea of qualitative physics with appreciation of the trends of the parameters that it describes. the idea of neuronal techniques with their adaptivity and their learning power and the idea of hybrid models integrating at the same time a combination of these three techniques. Control allows to master and guide the behaviour of a process in order to reach a given objective. This control concerns the evolution of some dimensions of the process (actions on input variables in order to subdue output or internal variables). The process control is generally established on the basis of intuitive causality schemes. For instance, starting from a cause, we deduce explicitly the effects that it engenders; this can be stated by the following rule : "If the state of the process is X and if I apply action V. then I observe that the process has the output Y ", where X is a set of information describing the behaviour of the process. However, if we know the effects produced, it is interesting to find the causes that produce them. This can be expressed by the following rule: "If the state of the process is X and if the process desires output Y, then apply V ". 893 To the intuitive stating schemes of the control. a set of knowledge is added : characterization of a desired behaviour; characterization of a real behaviour; taking into account of the dynamics of the process in terms of trends, prediction. analysis of the background; adaptation of the action policy to the dynamics of the behaviour of the process. Control has formed the subject of deep formalizing in the field of automation: open-loop control and its generalization, feed-back control, adaptative control, optimum control theory (Luzeaux, 1991). Other extensions of these models take into account the notion of uncertainty of the process control (Burg. 1988) (by integration of vague expressions into the model). Other more recent approaches. like those of qualitative physics (De Kleer. 1984). neuronal networks (McClelland, 1988) and symbolic control using formalisms of both qualitative physics, the fuzzy and classical automation (Foulloy, 1989) have allowed rich and pragmatic formulation of the process control. At present, control is defined in an approach that requires advanced research both in Artificial Intelligence and in Automation (Foulloy, 1990). In the works of Artificial Intelligence. the research focuses on the representation of knowledge of the process by means of : a structural description, a functional description, a description of the observations (phenomenological parameters) and a description of the actions acting on the process. The control is taken into charge by a scheme of reasoning that allows to pass from observation and