Model checking rational agents

Agent-oriented programming techniques seem appropriate for developing systems that operate in complex, dynamic, and unpredictable environments. We aim to address this requirement by developing model-checking techniques for the (automatic or semiautomatic) verification of rational-agent systems written in a logic-based agent-oriented programming language. Typically, developers apply model-checking techniques to abstract models of a system rather than the system implementation. Although this is important for detecting design errors at an early stage, developers might still introduce errors during coding. In contrast, developers can directly apply our model-checking techniques to systems implemented in an agent-oriented programming language, automatically verifying agent systems without the usual gap between design and implementation. We developed our techniques for AgentSpeak, a rational-agent programming language based on the AgentSpeak (L) abstract agent-oriented programming language. AgentSpeak shares many features of the agent-oriented programming paradigm. Similarly, we've developed techniques for automatically translating AgentSpeak programs into the model specification language of existing model-checking systems. In this way, we reduce the problem of verifying that an AgentSpeak system has certain BDI logic properties to a conventional LTL model-checking problem.