In classical AI planning, replanning strategies are used to reevaluate a plan during execution. For human-like agents, goal preferences and emotions play an important role in evaluating a plan's progress. However, most existing systems rely on predefined goal ordering and a static association between an event and its emotion-based utility calculation. During execution, utility of individual actions are used to trigger the replanning process. This approach assumes that a complete sequence of actions can be generated, preferences are known, are transitive, the mood-based utility of every action's outcome is known, and a replanning condition is well defined. This paper presents an alternative approach, one that does not make assumptions about the agent's or observer's omniscience about factors influencing decision making. Our approach recognizes the bounds that limit the agent and observer equally. To accommodate these limits, first, human-centric goal ranking are grounded in a domain-specific mapping to Maslow's hierarchy. Second, a new replanning condition is proposed with dynamically changing mood-based utility during plan execution.
Saving...