Atomic Intentions in Jason +

We consider interactions between atomic intentions and plan failures in the Jason BDI-based agent programming language. Atomic intentions allow the agent developer to control the execution of intentions in situations where a sequence of actions must be executed 'atomically' in order to ensure the success of a plan. However, while atomic intentions in Jason enforce mutual exclusion, they are not atomic operations in the sense understood in conventional programming or in databases, and failure of an atomic plan can leave the agent's belief and plan bases in an inconsistent state. In this paper we present a new approach to atomic intentions which provides a transactional 'all-or-nothing' semantics, and describe its implementation in a new version of Jason, Jason+. We argue that Jason+ offers a more predictable semantics for atomic plans in the face of plan failure and can reduce the load on the agent developer by automating simple cases of failure handing, leading to the development of more robust agent programs.