BDIggy: an agent archtecture to simulate human planning and human interaction

Keywords

Multi-agent systems, BDI agents, agent architectures, speech act theory, semantics of performatives, timed automata, cognitive modelling, human interaction, human planning, cooperative problem solving.

Publications

JFSMA03, Workshop@AAMAS03, IAT03, MFI05, IAT05, PhD, EuroCogSci07, AAMAS07, CEEMAS07, ICEIS08, RIA10.

Description

The BDIggy project aims at improving the design of multi-agent systems, thanks to the study and the modelling of human capabilities of planning and interaction. It is based on a psychological experiment where human subjects had to solve a planning problem with incomplete information.

The protocols collected during the experiment has been analysed from from the planning point of view and from the interaction point of view. Thus, a model of human planning and a model of human interaction are proposed.

The human planning model and the human interaction model are integrated homogeneously into a new agent architecture called BDIggy. Among existing agent models, the BDI (Belief, Desire, Intention) one offers an interesting framework to design deliberative agents able to plan and interact according to their mental states. Therefore, the BDIggy architecture is a merging of the Iggy system (a pre-existent human planning model) and the BDI architecture extended to a cooperative problem solving context.

The human interaction model included in BDIggy is twofold:

The utterance level and the discourse level are synchronized by a semantics of the performatives. This semantics summarizes the pre-conditions and the post-conditions that should be verified when receiving and sending a message, according to the knowledge of the agent.

The BDIggy architecture is validated with a Turing-like test by comparing the experimental protocols from the psychological experiment and the protocols generated by the BDIggy agents, during a simulation of the problem solving.

Approach and models have been extended to negotiation process.