An initial response to the OAS'03 challenge problem

Ian Dickinson, Michael J. WooldridgeOAS 2003 · OpenReview

Abstract

This paper presents an architectural approach to the OAS'03 travel planning challenge using the Nuin agent platform. By integrating Belief-Desire-Intention (BDI) reasoning with Semantic Web ontologies (DAML+OIL), the proposed system addresses challenges in heterogeneous vocabularies, vague user preferences, and proactive goal adoption in multi-agent environments. The work demonstrates how agents can move beyond simple data retrieval to perform sophisticated tasks like reconciling measurement units and ranking options based on fuzzy preferences.

Introduction

The OAS’03 Challenge Problem serves as a benchmark for testing agent-based systems in realistic, multi-actor travel scenarios. In this environment, a "travel agent" software must coordinate with various "supplier agents" (e.g., airlines, hotels) to fulfill a human client's request. This task is complicated by several real-world factors: heterogeneous vocabularies (e.g., miles vs. kilometers), vague user preferences (e.g., "high-quality" but "low-budget"), incomplete information held by self-interested suppliers, and the need for proactive behavior where agents infer goals from context.

The Nuin Platform and BDI Architecture

The authors address these challenges using the Nuin platform, built on the AgentSpeak(L) language and implemented in Java. The core of Nuin is a Belief-Desire-Intention (BDI) architecture, which models an agent’s mental state through beliefs, desires, and intentions.

The Nuin agent architecture, showing the integration of the agent core with BDI components and external service layers.

As shown in the architecture diagram, the system separates high-level reasoning from low-level service interaction via an abstract service adapter layer. This allows the agent core to interact with diverse services like JADE platforms or RSS translators while maintaining a consistent internal logic based on its BDI state.

Semantic Web Ontologies for Travel

A critical component of the system is the use of Semantic Web technologies, specifically DAML+OIL and RDF, to describe the travel domain. Unlike traditional databases, these ontologies support Description Logic (DL) reasoning.

The travel ontology class hierarchy developed for the OAS’03 challenge, illustrating the relationships between journeys, transport modes, and vehicles.

By utilizing the Jena toolkit, Nuin agents can reason through class hierarchies. For instance, an agent can automatically infer that a specific flight on an "A320" (a subclass of "AirbusPlane") constitutes "AirTravel." This hierarchical approach allows agents to handle varying levels of detail and reconcile different vocabularies through translator agents or normalized internal forms.

Key Insights and Qualitative Results

The research highlights several innovative approaches to agent behavior:

  • Proactive Goal Adoption: Agents can infer user goals from "inform" messages rather than waiting for explicit requests, enabling more collaborative behavior.
  • Preference Handling via Blank Nodes: RDF "blank nodes" are used as existential variables, allowing agents to treat user preferences (e.g., non-smoking rooms) as logical queries against potential solutions.
  • Logic-Based Ranking: The system demonstrates how to create a partial ordering of solutions using the transitive closure of preference rules (e.g., ranking 5-star hotels above 4-star).

The results are primarily architectural, proving the feasibility of mapping complex travel requirements to the Nuin platform's features and demonstrating interoperability between FIPA standards and Semantic Web technologies.

Conclusion and Future Work

While the Nuin platform successfully addresses many aspects of the OAS'03 challenge, several areas for improvement remain. Future work includes developing better "meta-level" reasoning to handle non-responsive suppliers ("monitoring silence") and integrating richer constraint languages to handle fuzzy temporal or spatial constraints (e.g., "around the 10th of July"). Additionally, the ethical and economic modeling of agent self-interest—balancing agency commissions against client needs—remains a significant open question in the field.

Summary updated

BibTeX
@inproceedings{dickinson2003initial,
  author = {Ian Dickinson and Michael J. Wooldridge},
  title = {{An initial response to the OAS'03 challenge problem}},
  booktitle = {OAS},
  year = {2003},
  url = {https://openreview.net/forum?id=uZXWlzixLm},
}
Figure