An alternative approach to modeling and analysis of agents’ behaviour is presented in this chapter. The agents and agent systems are understood here to be discrete-event systems (DES). The approach is based on the place/transition Petri nets (P/T PN) that yield both the suitable graphical or mathematical description of DES and the applicable means for testing the DES properties as well as for the synthesis of the agents’ behaviour. The reachability graph (RG) of the P/T PN-based model of the agent system and the space of feasible states are found. The RG adjacency matrix helps to form an auxiliary hypermodel in the space of the feasible states. State trajectories representing the actual interaction processes among agents are computed by means of the mutual intersection of both the straight-lined reachability tree (developed from a given initial state toward a prescribed terminal one) and the backtracking reachability tree (developed from the desired terminal state toward the initial one; however, oriented toward the terminal state). Control interferences are obtained on the base of the most suitable trajectory chosen from the set of feasible ones.
TopIntroduction
Behaviour of an agent in surroundings as well as among other agents in multi-agent systems (MAS) is one of the most important parts of the research in the intelligence integration. Agents are usually understood (Fonseca, Griss, & Letsinger, 2001) to be persistent (especially software, but not only software) entities that can perceive, reason, and act in their environment and communicate with other agents. Hence, MAS can be apprehended as a composition of collaborative agents working in shared environment. The agents together perform a more complex functionality. Communication enables the agents in MAS to exchange information. Thus, the agents can coordinate their actions and cooperate with each other. However, an important question arises here, namely: What communication mechanisms enhance the cooperation between communicating agents?
In general, the agent interaction is a specialized kind of the behaviour. Roughly speaking, the agent behaviour has both internal and external attributes. From the external point of view the agent is (Demazeau, 2003) a real or virtual entity that (i) evolves in an environment; (ii) is able to perceive this environment; (iii) is able to act in this environment; (iv) is able to communicate with other agents; and (v) exhibits an autonomous behaviour. On the other hand, from the internal point of view, the agent is a real or virtual entity that encompasses some local control in some of its perception, communication, knowledge acquisition, reasoning, decision, execution, and action processes. While the internal attributes characterize, rather, the agent inherent abilities, different external attributes of agents manifest themselves in different measures in a rather wide spectrum of MAS applications, like for example, computer-aided design, decision support, manufacturing systems, robotics and control, traffic management, network monitoring, telecommunications, e-commerce, enterprise modeling, society simulation, office and home automation, and so forth. Even (Demazeau, 2003), the applications in computer vision, natural language processing, spatial data handling and so forth, are known as well.
It is necessary to distinguish two groups of agents or agent societies, namely, human and artificial. The principle difference among them consists especially in the different internal abilities. These abilities are studied by many branches of sciences including those finding themselves out of the technical branches; for example, economy, sociology, psychology, and so forth. This chapter does not set itself these abilities as a goal of studies. It takes no account of the causes of them. Simply said, the internal behaviour happens and it is practically idle to consider how it happens. Here, the appearance of the internal abilities in the form of discrete events is important only. However, on the other hand, the external (i.e., inter-agent) behaviour is very important as to the quality of the communication or cooperation process in MAS. At the cooperation in MAS two principle characteristics of the agents are usually distinguished. Namely, either each agent is able to solve the whole problem but the use of many agents in parallel speeds up the problem solving, or the agents are specialized to solve different subproblems. While, in the former case, the cooperation consists of the purely physical (i.e., spatial or temporal) decomposition of the work between the agents, for example, each agent either solves a part of the problem or works for a given time, in the latter case each agent solves the problem for which it is specialized. However, a mix both of them seems to be more effective. Namely, it is very useful when an agent being free is able to substitute (at least partially) the activities of another agent in case of a failure or to help another agent asking for help (e.g., in case when it is not able to solve a problem).
As to the agent abilities, we can speak about cognitive and reactive agents. The cognitive agents are those that can form plans for their behaviours, whereas reactive agents are those that just have reflexes. Ferber (1999) showed how both approaches could converge in the end. Namely, one kind of research focuses on the building of individual intelligences whose communication is organised, whereas the other imagines very simple entities whose coordination emerges in time without the agents being conscious of it. However, in fact, a number of different schools of MAS persist, all coming from different theoretical backgrounds.