Automatic multi-platform implementation of AUML interaction protocols with the help of Petri nets
 |
 |
This project is based on the ability to represent a model of a multi-agent system described in Agent UML (AUML) through a particular Petri net called Multiagent Interaction Protocol Net (MIPNet). The Petri net can then be analyzed to identify properties that ensure the effective feasibility of the protocol.
The places of a MIPNet can be divided into sub-groups each associated with one of the entities that contribute to the implementation of the protocol and from these sub-groups is possible to extrapolate the behavior which the individual entities (agents) must follow in order to carry out properly the interaction.
The purpose is to provide a mechanism for automatic deployment to a real platform, building a system that ensures the proper implementation of the protocol to the agents who decide to take part. The analysis performed on the upstream MIPNet is independent of the realization, then, starting from the same data results from the analysis it is possible to create more than one mechanism of implementation, ideally one for each of the platforms on which we want to implement AUML models of multiagent systems.
Current status
Each model described by an AUML sequence diagram can be represented textually by combining fundamental units called Communication Act (CA) that represent the basic act of exchange of information between two different entities. A CA is nothing but a triple containing a source, a receiver and an identification of the type of information exchanged; the CAs are used to form interaction protocols by using logical operators similar to boolean operators, specifically AND, OR and XOR.
A parser, made using tuProlog, analyzes the interaction protocol in this way defined and creates a similar version that can be easily transformed into the associated MIPNet. The structure of MIPNets made from an AUML diagram, conditionally to a correct translation algorithm, is such so to guarantee liveness and the possibility to determine boundedness using LTL properties (the two properties that guarantee the feasibility of the interaction protocol). The MIPNet is expressed in Maude syntax to be able to use the features of the program for the analysis.
Using a correct modeling, it was possible to associate with the MIPNet constructed as described above, a representation compatible with the Jason syntax. The MIPNet was ideally divided into sub-networks, each one associated to the respective agent; such sub-networks were then transformed into plans that constitute the behaviour of the BDI agent. Using a technique similar to that used for the realization of the MIPNet from the AUML diagram, were then generated the source codes of agents, thus obtaining a mechanism for automatic implementation towards this platform.