A Theory of Intra-Agent Replanning
Kartik Talamadupula, David Edward Smith, William Cushing, Subbarao Kambhampati · 2013
david.smith @ nasa.gov When autonomous agents execute in the real world, the world state as well as the objectives may change from the agent’s original conception of those things. In such cases, the agent’s planning process must modify the existing plan to make it amenable to the new conditions, and to resume execution. The need for inter-agent replanning, in terms of commitments to other agents, is understood in the multi-agent systems com-munity. Such inter-agent replanning also motivates an intra-agent replanning problem for each individual agent. How-ever, the single-agent planning community has mostly limited its view of replanning to reducing the computational effort in-volved, by minimally perturbing the current plan structure to replan. This is not very appropriate as a general model for intra-agent replanning, which may consist of various tech-niques that are employed according to the scenario at hand. In this paper, we present a general replanning problem that is built on various types of replanning constraints. We show that these constraints can model different types of replanning, in-cluding the similarity-based approaches used in the past and sensitivity to commitments made to other agents. Finally, we show that partial satisfaction planning provides a good sub-strate for modeling this general replanning problem. 1