Using Observation to Select Complementary Goals

Satisficing Models

Volume

Satisficing Models

Contents

We present a goal-recognition based goal ordering strategy for an agent in a communication-free cooperative environment. Agents acting in such environments can improve coordination through observation and inference about the intentions of other agents. The central component of the strategy proposed in this paper is observation based goal recognition, that helps an agent understand what others are planning to do and accordingly select its goal to complement the behavior of the rest of the group. In cooperative as well as competitive environments an agent often needs to pursue goals not being pursued by others. Goal recognition and replanning are usually time consuming, thus they should be used only when it is necessary. Some of the questions we are addressing are: how often and when an agent needs to call the goal recognizer; how far ahead should an agent plan in a dynamic environment and at what level of detail should this plan be generated; when does it needs to replan? We also investigated planning and plan recognition on multiple levels of abstraction.