HANA: A Human-Aware Negotiation Architecture

摘要

In this paper we propose HANA, a software architecture for agents that need to bilaterally negotiate joint plans of action in realistic scenarios. These negotiations may involve humans and are repeated along time. The architecture is based on a BDI model that represents the uncertainty on the environment as graded beliefs, desires and intentions. The architecture is modular and can easily be extended by incorporating different models (e.g. trust, intimacy, personality, normative…) that update the set of beliefs, desires or intentions. The architecture is dynamic as it monitors the environment and updates the beliefs accordingly. We introduce an innovative search&negotiation method that facilitates HANA agents to cope with huge spaces of joint plans. This method implements an anytime search algorithm that generates partial plans to feed the negotiation process. At the same time the negotiation guides the search towards joint plans that are more likely to be accepted.