History

Object Networks (ONs) was introduced by Gudwin [Gudwin 1996] as a computational tool for representation and processing of knowledge structures under The Computational Semiotics paradigm [Gudwin 1996, Gudwin & Gomide 1997].

Basically, an ONs is formed by a group of mathematical objects, that are entities that can present states and transformation functions that act in these states, a topology constituted of places and arcs interconnected and a mechanism that governs the behavior of the objects. The behavior, that is, which objects discharge and with whom them interact, is implemented like this by the called selection function (SF), which is defined for all objects of the network. The notion of SF is a generic representation of behavior the objects given a group of restrictions, so that an ONs can model sophisticated and adaptable behaviors because an SF can be defined in the time.

 

For each object, the construction of selection functions is responsibility of the model designer. Thus, in this context, becomes important the study of auxiliary tools that allow a more direct application of ONs model in specific domains. An example of ONs can be observed below. This is an Object Networks to model the learning of ONs using a Genetic Algorithm.

 

In the initial phase of the research of ONs, very difficulties in effective implementations of simulation models were founded due to the lack of generic software libraries, because all problems used specific implementations, restricted to its own model and environment of programming. The high level of abstraction of ONs acts as a factor that can elevate the complexity of the implementations considerably.

 

One of the more important models implemented under the paradigm ONs comes to be a control system of an autonomous vehicle [Gudwin 1996]. This system presents a great amount of places and different classes, as well as an amount relatively big of resources.

 

In this sense, a first step was presented in [Guerrero et. al. 1999], based on the proposition of an algorithm for the determination of some components of SF. Later on, in [Guerrero 2000], this idea was then integrated into the original model of Object Networks, creating Agent Networks (ANs).

The ANs is a special type of Agent Systems with topological restrictions given by the addition of two types of passive entities: places and arcs.

 

Starting from these information the algorithm BMSA [Guerrero 2000] determines which initials proposition will be exactly executed. The nomenclature "Agent Networks" reflects the autonomous behavior of an active object, because in this case, each one of them possesses a certain autonomy level, dictated by the own interest.

For the development of Object Networks(ONs) and Agent Networks(ANs) a generic environment of ONTOOL was created [Guerrero 2000].

 

The environment ONTOOL-1 represents a progress in the study of ONs and ANs for allow the creation of models relatively sophisticated without necessity of the extensive coding and verification. This system presents the following characteristic:

    (a) attribution of Thread of execution to each place of the network 

    (b) support to external classes

    (c) algorithm of determination of selection function BMSA  (Best Matching Search Algorithm):   

 

    (d) direct implementation of classes in Java language.

The system ONTOOL-1 uses a model client/server for the integration of its three modules, see below. In this way, the modules of the system could be distributed in different machines by the network.

Many examples were developed using the ONTOOL-1, for example, Colored Petri Network (CPN), solution of the Traveling Salesman Problem using genetic algorithms (Agent Networks for TSP), simple sailing of an autonomous vehicle using a Fuzzy controller, the eating philosophers problem [Guerrero 2000] and for the control of the autonomous vehicle [Suárez 2000], see below:

With the natural evolution of the studies in Agent Networks and the attempt of extending the possible fields of application, new requirements for the system ONTOOL were originated. The architecture composed by three components was limited by hindering the application in some complex problems (with more than a simulation nucleus), to hinder management of the modules (due to the specific communication protocols) and expansion of the functionalities (it supports to modular agent networks).   

Thus, a new version for the environment of development of Agent Networks was created, looking for progresses in several aspects, from the graphic interface to the support of the network ONTOOL-2 [Gomes 2000].

The system ONTOOL-2 introduces new characteristics: 

The graphic interface had you vary modifications, the most important are simultaneously the capacity of edition of properties of multiple elements, as arcs, places, super-places, code of the implementations of the class. Also is included an editor of classes in the graphic format, in agreement with a representation of the formal agent, Figura 5.3 (EPS)

 

 

 

 

 

Another important characteristic is the edition of sensitive mechanisms of help to the context, Figura 5.4 (EPS)

 

 

 

 

 

The illustration below exhibe in a simplified form the cycle of typical development inside of the enviroment ONTOOL. Initially the user elaborates the model computational model starting from a previous conceptual model, compiled later. In this stage the code that indeed implements the model is generated in the form of a collection of  Java specific classes to the suitable model and the library of simulation MTON.  

An important characteristic in this architecture is the existence of two types of classes: