README.md

Proactive Agent Behaviour in Dynamic Distributed Constraint Optimisation Problems.

This is the codes accompanying the paper with the title above. There are three parts to this project:

The Robocup Rescue Simulation Server

The RCRS is the simulation environment used for evaluating the proposed approach in the study. Follow the official repo to set it up locally.

Agent Communication

The communication layer used by the agents in the experiments rely on the AMQP RabbitMQ broker. Follow the official installation guide to set the broker up locally. We found it easier to use the Docker image of RabbitMQ for our experiments. We used the Python client library (Pika) in this repo.

The Agents Implementation

The dependencies of this project can be installed by running:

$ pip install -r requirements.txt

Once the RCRS server and RabbitMQ broker have been started, the agents program in this codebase can be started by running:

$ python launch.py -fb 12 -fs 1

The above command starts 12 fire brigade agents and 1 fire station.

See rcrs_ddcop.core.fireBrigadeAgent for the fire brigade agent program. Each fire brigade agent composes a core.bdi_agent.BDIAgent program that is in charge of spinnig up the DCOP algorithm, the dynamic graph algorithm, the information sharing algorithm, experience buffer, and other components used for agent reasoning.

The algorithms.dcop.DCOP class is the parent class of the DCOP algorithms used in the experiments. It creates the XGBoostTrainer which trains the local agent model on the shared experiences in the buffer.