Planning in dynamic worlds: More than external events

Hardly any environment is a static domain in which an agent is the only one who is changing states of the world. Thus, many events occur that are not a direct consequence of an agent’s actions. Besides events that occur in full independence of an agent, there are events that can be influenced indirectly, e.g., by asking other agents to perform specific tasks. Even in the absence of other agents, an action by an agent can lead to a complex effect chain. Explicitly reasoning about such indirect consequences of actions is indispensable in nearly all real-world domains. Action planning systems to drive an intelligent agent, however, do not incorporate concepts to handle such world dynamics. An action usually has a definite effect — without a possibility of further indirect consequences. Some planning systems allow for the occurrence of external events, but do not enable relations to an agent’s actions. In this paper, we propose a solution to this shortcoming of planning systems by integrating a kind of enhanced rule-based system. Using this approach, a planning system can reason about indirect consequences and exploit these external mechanics in order to achieve its goals. The enhancements are demonstrated from the perspective of a constraint-based planning system based on local search