In: Proc. IEEE Int. Conf. on Systems, Man, and Cybernetics (SMC'99), 12-15 October 1999, Tokyo, Japan, Vol. 5, pages 143-148. 1999.
Abstract: Performance evaluation of multi-agent robotic systems requires a great amount of simulations in order to check the emergence of the required collective behavior. This paper proposes to model such systems by the stochastic Petri nets. In this way, it is shown how to use the powerful tools of Markov chains analysis for computing quickly the evolution of states of the multi-agent systems and of the environment. A simple example illustrates the presentations, and the theoretical results are compared to simulations. It concerns a homogeneous population of Tom-Thumb robots collecting samples situated in some unknown places (e.g., mines) and bringing them to a home base. The Markovian model which is able to provide the system's state at any time, is solved very quickly, while simulations are long and tedious when the performance must be evaluated as a function of various parameters: number of robots, number of mines, number of samples, etc.
Keywords: multi-agent systems, performance evaluation, stochastic Petri nets.