A Failure/Fault Diagnoser Model for Autonomous Agents under Presence of Disturbances

In this paper, we provide our preliminary results on failure/fault analysis for an autonomous agent whose behavior is influenced by disturbances from its environment. It is assumed that failures are unobservable and only detected through some observable symptoms. Furthermore, faults may be observable directly, but the conditions leading to them are unknown a priori. Both the agent and its environment may contribute to any given failure or fault. The main results of this paper relate to the development of a diagnoser model which is built by a parallel composition of Petri net models of the underlying components. This model includes both the normative and disruptive behavior of the agent and its perception of the environment. The uncontrollability of the environment leads to a non-deterministic diagnosis behavior. The likelihood of each possible state of the agent´s diagnoser is updated every time that a new controllable action is taken or new sensory information is received. At each updating, if the likelihood of a failure or fault state is higher than a specified threshold, then an alarm is raised indicating a disruptive event.