Agent deliberation via forward and backward chaining in linear logic

Agent systems are designed to work in complex dynamic environments, which requires an agent to repeatedly deliberate over its choice of actions. A common way to achieve this is to use agent architectures based on the Belief-Desire-Intention (BDI) model, in which an agent continuously deliberates over the best way to achieve its goals in the current environment. In this paper we explore how a BDI approach can be implemented in Lygon, a logic programming language based on linear logic. In particular, we show how backward and forward chaining techniques can be used to provide proactive and reactive agent behaviours. We discuss some extensions to Lygon which allow us to use abduction techniques to generate plans to achieve a given goal, as well as an addition to the syntax of Lygon which greatly simplifies the specification of a sequence of goals to be achieved. We also show how a simple addition to the backward chaining process allows us to specify proactive checking of maintenance goals.