Motivations and Goal-Directed Autonomy

As we move towards a richer model of computation based on interactions among autonomous agents in lieu of the more traditional numeric batch processing model, it is necessary to devise techniques and abstractions for the development of such autonomous agents. The ability to generate, reason about, and select the means to achieve explicit goals is critical for autonomous agents, the so called goal-driven autonomy. Reasoning about goals demands a suitable abstraction for meta-level reasoning, so that an agent not only reasons about the actions it needs to take, but also about the viability of (and interaction among) such goals. In this paper we discuss the use of motivations as an abstraction of metareasoning for goal-driven autonomous agents. We review recent developments in motivated agent architectures and discuss possible directions for future research.