We presented a model for autonomy-based multi-agent systems and its prerequisites for parallelization. We built a coordination platform based on STL's coordination model on top of the existing PT-PVM platform [Krone et al. 1996]. We sketched a preliminary STL-based distributed implementation of our multi-agent model applied to a collective robotics simulation. STL showed its power and demonstrated its appropriateness for coordinating a generic class of autonomous agents, whose most critical constraint is the preservation of temporality by dismissing coordination mechanisms exclusively embedded for purpose of implementation.
There are two major outcomes to this work. First, as autonomous agents' systems are aimed at addressing problems which are naturally distributed, our coordination platform provides a user the possibility to have an actual distributed implementation and therefore to benefit from the numerous advantages of distributed systems, making this work a step forward in the Autonomous Agents community. Secondly, as the generic patterns of coordination for autonomy-based multi-agent implementations are embedded within the platform, a user can quite easily develop new applications (e.g. by changing the type of autonomy of the agents, the type of environment), insofar they comply with the generic model.
Future works are as follows. First, a graphical user interface will be developed in order to facilitate the specification of the coordination part of a distributed application. Secondly, the basic mechanisms of STL have to be enhanced in order to simplify the expression of the coordination and to establish a well defined semantics. Thirdly, the implementation of STL is still to be carried on in two ways: as a separate coordination language and as a coordination library. Fourthly, the STL coordination model is still to be extended in order to encompass as many as possible generic patterns of coordination, yielding in STL templates at disposal for general purpose implementations. In our study, we focused only on operational autonomy in an engineering approach. Future works will consist in studying behavioral autonomy through learning approaches.