An Action Selection Architecture for Autonomous Virtual Agents
Résumé
Many day-to-day applications involve autonomous agents. When designing autonomous agents, the problem of selecting actions must be considered, as it governs decision making at all times. In this paper we describe how we designed intelligent virtual cognitive agents representing operators performing tasks in a high-risk plant. They must respond to expected as well as unexpected events. The reasoning system controlling the agent behavior is designed to exhibit human behavior. Agents must be able to plan their actions according to their perception, their beliefs and their goals. We developed a planning system that interleaves plan construction and plan execution. The new planner, called AATP, produces a plan according to agent goal.