Abstract. Global Abduction (GA) is a recently proposed logical formalism for agent oriented programming which allows an agent to collect information about the world and update this in a nonmonotonic way when changes in the world are observed. A distinct feature of Global Abduction is that in case the agent needs to give up one plan, it may start a new one, or continue a suspended plan, while its beliefs learned about the world in the failed attempts persist. This paper describes an implementation of GA in the high-level language of Constraint Handling Rules (CHR). It appears to be a first attempt to a full implementation of GA, which also confirms CHR as a powerful meta-programming language for advanced reasoning. The construction gives rise a discussion of important issues of the semantics and pragmatics of Global Abduction, leading to proposals for a specific procedural semantics and architecture that seem well suited for real-time applications. 1

Standard procedures of logic programming do not take into account program updates. If the given program is updated during or after the evaluation of a query, the answer to the query is not correct any longer. This paper shows how to use Dynamic SLDNF (DSLDNF), a new extension of SLDNF, which works under changing logic programs in order to construct plans for robot navigation in dynamic environments. 1 Introduction In most literature on logic programming, the design of a proof procedure does not take into account the dynamic nature of programs. If this kind of proof procedure is used, every time the program is updated, the query has to be re-evaluated from scratch. However, it is easy to imagine that if the program is changed only a little bit, the old proof is still valid in many cases. Dynamic SLDNF (DSLDNF) [ Hayashi, 1999 ] , an extension of SLDNF 1 , was developed so that it achieves correct answers to queries under changing logic programs. DSLDNF can suspend query eva...

In a dynamic environment, even if an agent makes a plan to obtain a goal, the environment might change while the agent is executing the plan. In that case, the plan, which was initially valid when it was made, might later become invalid. Furthermore, in the process of replanning, it is necessary to take into account the side e#ects of actions already executed . To solve this problem, we have previously presented an agent life cycle that interleaves HTN planning, action execution, knowledge updates, and plan modification. In that agent life cycle, the plans are always kept valid according to the most recent knowledge and situation.