We study agents situated in partially observable environments, who do not have sufficient resources to create conformant (complete) plans. Instead, they create plans which are conditional and partial, execute or simulate them, and learn from experience to evaluate their quality. Our agents employ an incomplete symbolic deduction system based on Active Logic and Situation Calculus for reasoning about actions and their consequences. An Inductive Logic Programming algorithm generalises observations and deduced knowledge so that the agents can choose the best plan for execution.

We describe an architecture which allows ideas and solutions from several sub-fields of Artificial Intelligence to be joined together in a controlled and manageable way. In our opinion, no situated agent can achieve true rationality without using at least logical reasoning and learning. In practice, it is clear that pure logicis not able to cope with all the requirements put on reasoning, thus more domain-specific solutions, like planners, are also necessary. Finally, any realistic agentneeds a reactive module to meet demands of dynamic environments. Our architecture is designed in such a way that those three elements interact in order to complement each other’s weaknesses and reinforce each other’s strengths.