Overview
Distributed Artificial Intelligence expands distributed systems by embedding learning, cooperation, and autonomy into each agent. Initial systems used centralized Distributed Problem Solving (DPS) with shared goals, but later Multi-Agent Systems (MAS) leaned toward decentralized planning. The thesis argues that MAS still benefits from a guiding DPS layer by improving how resources and information are allocated.
Research Objectives
The goal is to measure whether MAS agents controlled by a DPS-style manager can post higher stability and efficiency than MAS alone. The plan: half the project time is dedicated to reviewing DPS and MAS implementations, followed by concurrent research and development. After building the experiment, the final phase compares the DPS-directed MAS against the null hypothesis across varied conditions.
Approach
The approach supplies a network of agents running MAS decisions with partial global planning heuristics. When the DPS controller approves actions, agents get updated resources or information. Frequent professor meetings steer research direction, while experiments rely on the applicant's experience with distributed-parallel systems to build the agent ensemble.
Applications
Proposed deployments include forestry and marine management: MAS can schedule land use based on sensor data, maximize yield while allowing rest periods, and help plan conservation efforts. Marine scenarios cover route allocation with tidal and pollution data. These applications demonstrate how DPS-informed MAS can automate preventative actions for national parks and shipping sustainability.
References
The write-up cites foundational MAS planning literature: Clement & Weerdt (2009) on planning, Durfee (1994) on DPS vs MAS, Elliot (2005) on agent semantics, and Yeoh & Yokoo (2012) on distributed problem solving.