Detecting commitment abandonment by monitoring sub-optimal steps during plan execution

Assessing whether an agent has abandoned a goal is important when multiple agents are trying to achieve joint goals, or when agents commit to achieving goals for each other. Making such an inference for a single goal by observing only plan traces is not a trivial task because agents often deviate from the optimal plans for various reasons, including the pursuit of multiple goals or the inability to act optimally. In this paper, we develop an approach that uses plan optimality monitoring techniques to determine whether an agent will honour a commitment. Specifically, to determine commitment abandonment, we use these techniques with planning fact partitions (e.g., dead-ends). Our approach is domain-independent and detects commitment abandonment in nearly all cases.