Diploma Thesis
| Author | Andrzej Walczak |
| Supervisors | Winfried Lamersdorf, Daniel Moldt |
| Tutors | Lars Braubach, Alexander Pokahr |
| Title | Planning and the Belief-Desire-Intention Model of Agency |
| Abgabe am | 06.04.2005 |
| Abstract | Artificial intelligence researches the field of planning as a process able to provide agents with a course of action guiding them towards their aims. Agents incorporating the belief-desire-intention model of agency are generally equipped with a library of plans, avoiding the burden of planning at runtime. Nevertheless, there is a need for planning techniques within BDI agents, in order to cope with problems, for which it is difficult to create generic plans in advance. It needs to be investigated in theory and practice, what techniques are appropriate for this purpose. How to integrate planning with the BDI model of agency and how well does planning perform in multi-agent environments. It is also interesting if a successful planner could be implemented with means of Javatm. The planning community has developed many concepts with respect to this topic. Simple state space approaches, hierarchical task networks, partial order and deductive planning techniques can be used for the purpose of this diploma thesis. To integrate a planner with a BDI system an approach may be pursued where the planner controls the underlying BDI system or it uses BDI goals and plans to compose them into new plans in a dynamic way. On the other hand, the BDI engine could also trigger the planner, to create plans composed of specially crafted operators. The aim of this diploma thesis project is to implement a planner in Javatm language that uses Java(TM) as its representation for planning concepts and to adapt and integrate this planner into the Jadex BDI reasoning engine. The applicability of modern planning techniques to multi-agent systems should also be proved. The planning techniques are investigated on planners used in the practice. They are compared to each other on the basis of efficiency, methods used and features supported. Most successful concepts are taken to design and implement a planner that would fit into a BDI agent architecture, in particular into the Jadex BDI agent system. The overall approach is evaluated on small examples. |
| Document |  |
| Other formats |  |
