Model Checking Artificial Intelligence Based Planners: Even the Best Laid Plans Must Be Verified

摘要

Automated planning systems (APS) are gaining acceptance for use on NASA missions as evidenced by APS flown On missions such as Orbiter and Deep Space 1 both of which were commanded by onboard planning systems. The planning system takes high level goals and expands them onboard into a detailed of action fiat the spacecraft executes. The system must be verified to ensure that the automatically generated plans achieve the goals as expected and do not generate actions that would harm the spacecraft or mission. These systems are typically tested using empirical methods. Formal methods, such as model checking, offer exhaustive or measurable test coverage which leads to much greater confidence in correctness. This paper describes a formal method based on the SPIN model checker. This method guarantees that possible plans meet certain desirable properties. We express the input model in Promela, the language of SPIN and express the properties of desirable plans formally.