Extending BDI Agents with Robust Program Execution, Adaptive Plan Library, and Efficient Intention Progression

  • Mengwei Xu

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)


The Belief-Desire-Intention (BDI) architecture, where agents are modelled based on their (B)eliefs, (D)esires, and (I)ntentions, provides a practical approach to developing intelligent agent systems. These agents operates by context sensitive expansion of plans, thus allowing fast reasoning cycle. However, the practical capability of BDI agents can still remain limited due to the lack of abilities to handling execution failure, adapting to the environment, and pursuing multiple intentions correctly and efficiently. In this thesis, we will address these issues in the following ways.
Firstly, we introduce a novel operational semantics for incorporating First-principles Planning (FPP) to recover execution failure by generating new plans when no alternative pre-defined plan exists or worked. Such a semantics provides a detailed specification of the appropriate operational behaviour when FPP is pursued, succeeded or failed, suspended, or resumed in BDI. Therefore, the robustness of a BDI agent can be substantially improved when facing unforeseen situations.
Secondly, we advance the state-of-the-art in BDI agent systems by proposing a plan library evolution architecture with mechanisms to incorporate new plans (plan expansion) and drop old/unsuitable plan (plan contraction) to adapt to changes in a realistic environment. Such a proposal follows a principle approach to define plan library expansion and contraction operators, motivated by postulates that clearly highlight the underlying assumptions, and quantified by decision-support measure information. Therefore, the adaptivity of BDI can be improved for a fast-changing environment.
Thirdly, we provide a theoretical framework where FPP is employed to manage the intention interleaving in an automated fashion. Such a framework employs FPP to plan ahead to not only avoid the potential negative intention interactions, but also capitalise on their positive interactions (i.e. overlapping sub-intentions). As a benefit, the achievability of intentions (i.e. a correct execution) is guaranteed, and the overall cost of intentions execution is reduced (i.e. an efficient execution).
Date of Award26 Nov 2020
Original languageEnglish
Awarding Institution
  • University of Bristol
SupervisorWeiru Liu (Supervisor)


  • Autonomous Agents
  • BDI Agents
  • Planning

Cite this