Autonomous agents architectures and algorithms in ¯exible manufacturing systems LUDOVICA ADACHER 1 , ALESSANDRO AGNETIS 2 and CARLO MELONI 1 1 Dipartimento di Informatica e Automazione, Universita Á di Roma Tre, via della Vasca Navale, 79-00146, Roma, Italy E-mail: adacher@dia.uniroma3.it or meloni@dia.uniroma3.it 2 Dipartimento di Ingegneria dell'Informazione, Universita Á di Siena, via Roma, 56-53100, Siena, Italy E-mail: agnetis@dii.unisi.it Received April 1999 and accepted November 1999 This paper investigates possible implementations of the autonomous agents concept in ¯exible manufacturing control. The im- plementation issues and the eectiveness of dierent control architectures and algorithms are analyzed by means of a simulation model of a ¯exible job shop. Extensive experimental results are reported, allowing the evaluation of the trade-o between the degree of autonomy and system performance. 1. Introduction Autonomous agents (AA) are becoming increasingly popular in ®elds related to computer applications, although this popularity is often disguised by the use of dierent concepts. In manufacturing, telecommunica- tions, medicine and public administration dierent enti- ties such as intermediate providers, departments, wards or even ®nal users may have sucient freedom to orga- nize their own activity within a general framework, without depending on external in¯uences (Jennings and Wooldridge, 1998). In software design and development, AA architectures provide a modular framework in which the actors recall very closely the objects of Object- Oriented Programming (OOP) (Booch, 1994; Franklin and Graesser, 1997; Pinedo and Yen, 1997). In all these cases, the underlying concept is that a project or a process can be regarded as being the result of interactions between several dierent subjects, rather than the result of a single centralized decision maker. Each individual acts in order to pursue his/her/its own objectives, which may be sometimes expressed by means of a mathematical objective function. Generally speaking, for the individuals to carry out their tasks in a feasible and pro®table way for the overall system, they will have to cooperate and negotiate to some extent. Cooperation means that two individuals may realize that taking a common action may turn out to be convenient for both of them. Negotiation refers to the fact that one individual may accept losing something in exchange for gaining something else. With respect to centralized decision- making, autonomous agents typically oer such advan- tages as management simplicity, ¯exibility, modularity and ease of monitoring (Malone, 1987). Manufacturing is a natural setting in which the autonomous agents concept may yield considerable pay- os. In the organization of a factory, it is natural to identify those agents with elements that have a certain amount of behavioral autonomy, be it a department, an oce, a machining center, a single worker or a workpart. Of course, several questions immediately arise, which do not necessarily call for a unique answer. What are sensi- ble objectives for the agents? What do autonomy, coop- eration and negotiation mean in this context? What does the performance of the system depend on? While studies exist comparing centralized versus distributed control systems (Brandimarte, 1993; Uzsoy and Ovacik, 1997), to date we know of only a few quantitative studies that consider the use of autonomous agents in manufacturing. In fact, the AA concept goes one step beyond distributed control (MuÈller, 1997). Sequencing rules (Morton and Pentico, 1993) are widely used in the manufacturing practice, due to their simplicity and relative eciency. However, a traditional shop ¯oor driven by sequencing rules cannot be considered an AA structure, for two main reasons. First, the workload of a machining center is typically assigned by an external dispatcher, and usually machining centers cannot react to it. Second, no real cooperation or negotiation takes place among the agents. The AA structure calls for a dierent idea of establishing the workload of the centers. This assignment may be the result of a bargaining phase in which the task to be 0740-817X Ó 2000 ``IIE'' IIE Transactions (2000) 32, 941±951