Programming agent deliberation using procedural reflection

A key advantage of BDI-based approaches to agent programming, is that agents can deliberate about which course of action to adopt to achieve a goal or respond to an event. However, while state-of-the-art BDI-based agent programming languages allow the programmer to specify the context(s) in which a particular plan is applicable, they are typically limited to a single, hard-coded, deliberation strategy for all task environments. In this paper, we present an alternative approach, in which an agent’s deliberation strategy forms part of the agent program. We show how both conventional agent programs and the agent’s deliberation strategy can be encoded in the agent programming language meta-APL. Key steps in the deliberation cycle of meta-APL are reflected in the state of the agent and can be queried and updated by meta-APL rules, allowing application-specific BDI deliberation strategies to be programmed in a straightforward way. To illustrate the flexibility of meta-APL, we show how three typical BDI deliberation strategies can be programmed using meta-APL rules. We then show how meta-APL can used to program a simple adaptive deliberation strategy that avoids interference between intentions