Failure Analysis for Adaptive Autonomous Agents using Petri Nets

Adaptive autonomous (AA) agents are able to make their own decisions on when and with whom to share their autonomy based on their states. Whereas dependability gives evidence on whether a system, (e.g. an agent team), and its provided services are to be trusted. In this paper, an initial analysis on AA agents with respect to dependability is conducted. Firstly, AA is modeled through a pairwise relationship called willingness of agents to interact, i.e. to ask for and give assistance. Secondly, dependability is evaluated by considering solely the reliability attribute, which presents the continuity of correct services. The failure analysis is realized by modeling the agents through Petri Nets. Simulation results indicate that agents drop slightly more tasks when they are more willing to interact than otherwise, especially when the fail-rate of individual agents increases. Conclusively, the willingness should be tweaked such that there is compromise between performance and helpfulness.