Composing high-level plans for declarative agent programming

Research on practical models of autonomous agents has largely focused on a procedural view of goal achievement. This allows for efficient implementations, but prevents an agent from reasoning about alternative courses of action for the achievement of its design objectives. In this paper we show how a procedural agent model can be modified to allow an agent to compose existing plans into new ones at runtime to achieve desired world states. This new agent model can be used to implement a declarative goals interpreter, since it allows designers to specify only the desired world states in addition to an agent’s basic capabilities, enhancing the agent’s ability to deal with failures. Moreover our approach allows the new plans to be included in the plan library, effectively enabling the agent to improve its runtime performance over time.