Welcome to the AbstractSwarm Multi-Agent Logistics Competition!


News


Aims and Scope

A long-term goal of Artificial Intelligence (AI) research is the development of “general” intelligence which is able to solve a broad variety of different problems. To further approach this extensive goal, problem scenarios from the broad field of logistics seem to serve as eligible test environments, that can be extremely diverse, highly dynamic and variable of size. Furthermore, many logistics-related problems (e. g., in hospital logistics) have a good accessibility and are easy to comprehend, but hard to solve.

The main goal of the competition is to stimulate AI/CI research in the context of multi-agent and swarm systems as cooperative problem solvers for a priori unknown problems, with a focus on logistics—especially for hospital process optimization in medical informatics.

For this purpose, the AbstractSwarm simulation system will be used, which is currently developed at the Medical Informatics department of the Institute of Medical Biostatistics, Epidemiology and Informatics (IMBEI) at the University Medical Center of the Johannes Gutenberg University Mainz, Germany.

The AbstractSwarm Multi-Agent Logistics Competition calls for applying results from AI/CI, multi-agent/swarm systems and related research to logistics, especially focusing on the cooperativity of agents.

This is the first round of the competition, which runs jointly at IEEE CEC 2021 and GECCO 2021, to be held in June/July. The announcement of the winners as well as all other results will be presented both at IEEE CEC 2021 and GECCO 2021 (exact time and date to be announced).


The Task

In the AbstractSwarm Multi-Agent Logistics Competition, participants must develop agents that are able to cooperatively solve different smaller a priori unknown logistics problems. A logistics problem is given as a graph containing agents (round components) and stations (squared components). Agents have to visit their connected stations at the right time to prevent other agents from having idle time. An agent can interact with the graph

While simulating a scenario, a Gantt-chart is created according to the decisions of all agents. The total number of idle time of all agents in the resulting chart is used as evaluation criterion.

Competition

Getting Started

  1. Scan through this paper about the syntax and semantics of the logisics graphs used for this competition—you don't have to read it entirely, but keep it, since it might be helpful from time to time.
  2. Download the latest realease build of the AbstractSwarm simulation system from the releases page. For further instructions on how to setup and run AbstractSwarm refer to the project site.
  3. Make yourself familiar with the system by loading one of the provided example graphs and simulate it with one of the provided example agents (you can also follow the instructions of the first tutorial video).
  4. Have a look at the implementation of the provided agents and start developing your own agents (see the second tutorial video for details on this).


Competition Rules


Videos/Tutorials


Introductory Presentation Video of the Competition Shown at GECCO 2021)


Tutorial 1: Core Ideas, Loading & Running Scenarios


Tutorial 2: How to Create Agents


Tutorial 3: Running Simulations from the Command Line

This brief tutorial explains how to run AbtractSwarm simulations from command line:
  1. Open a command line in the AbtractSwarm main folder (where AbstractSwarm.exe is located).
  2. Type the command:

    AbstractSwarm SCENARIO_PATH AGENT_NAME RUNS

    where SCENARIO_PATH is the (relative) path to the scenario, AGENT_NAME is the name of the folder of your agent implementation in the Agents subfolder and RUNS is the number of runs to be simulated.
  3. While the simulation runs, the remaining runs are counted down and two files are written: events.log and timetables.log.
    The file events.log contains all entering and leaving events of all agents for each time unit in which they occurred.
    In the file timetables.log, each line represents the sum of all waiting times (i. e., the unused timeunits) after a complete run. Empty lines indicate that the agents where not able to find a solution in this run.
  4. Once all runs are finished, the sum of all waiting times (i. e., the unused timeunits) of the best timetable found so far is shown on the command line.


Test Scenarios

Up to ten small episodic example scenarios of different (simplified) logistics and related problems will be provided here for download, which you can load using the AbstractSwarm system for testing or training your agents.

Treatments Treatments Ten patients have to get three different treatments each. The treatment locations have different amounts of space and have a distance of five units to each other. Spatio-temporal scheduling problem.
Ward Round WardRound A doctor has to do a ward round to visit four patient rooms in a hallway. A nurse has to do some office work, but must be available every time the doctor visits a room. Spatio-temporal coordination problem with only two agents.
X-Ray Scheduling XRayScheduling Twenty patients that have to be x-rayed must be distributed to three x-ray stations. Multi-agent scheduling problem.
X-Ray Scheduling 2 XRayScheduling2 Same as X-Ray Scheduling with distances among the three x-ray stations being additionally modeled. Multi-agent scheduling problem with spatial dependencies.
Crossroads Crossroads Trucks on two different roads (a slow one and a fast one) have to get from a starting point to a destination. The roads are crossing in the middle, where an agent has to control the traffic lights to optimize the traffic flow (colored in green). Traffic flow control problem with directed spatial dependencies and a cell-based character (where stations represent the cells of the roads).
Storage Storage The best way of storing ten products in three shelves of a linearly organized storehouse must be found here (where the products are represented by agents). The solution is obviously to store all ten products in the first shelf (colored in yellow), since it is located at the entrance of the storehouse and thus distances for storing can be minimized. Storage problem with sparse rewards (agents only receive better rewards if all agents decide to avoid one of the shelves located in the back).
Delivery Delivery Three deliverers must coordinate each other to provide one delivery for each of the six customers in a district. The delivers start at the bottom left customer (colored in white). Multi-agent delivery problem.
(More to come...)


Submission

Please submit your agent implementation by email to apeldoorn@uni-mainz.de, either as a zipped attachment or by providing a link to the corresponding .zip file in your repository, not later than:

15th of May, 2021

The .zip file should contain the complete agent implementation folder, including the AbstractSwarmAgentInterface.java file together with all other needed files (further .java files, .jar files of needed external libraries, files to be loaded, and the like).

Optionally, you can additionally submit a competition entry describing your agent implementation (max. two pages) to the GECCO 2021 conference, not later than:

12th of April, 2021

For details see the GECCO 2021 call for competition entries. A preview of the submission form can be found here.


Awards

We are planning to apply again for funding to award the winners of the next year's round.


Citation

If you consider submitting a competition entry or full paper related to the AbstractSwarm Multi-Agent Logistics Competition, you can cite AbstractSwarm as follows:

Apeldoorn, D.: AbstractSwarm – A Generic Graphical Modeling Language for Multi-Agent Systems. In: Klusch, M., Thimm, M., Paprzycki, M. (eds.) Multiagent System Technologies. LNCS, vol. 8076, pp. 180–192. Springer, Berlin Heidelberg, 2013.

and/or refer to this website.


Organizers

Apeldoorn, Daan (University Medical Center of the Johannes Gutenberg University Mainz, IMBEI Medical Informatics)
Dockhorn, Alexander (Queen Mary University of London)
Hadidi, Lars (University Medical Center of the Johannes Gutenberg University Mainz, IMBEI Medical Informatics)
Panholzer, Torsten (University Medical Center of the Johannes Gutenberg University Mainz, IMBEI Medical Informatics)


Questions?

If you have further questions, please contact the competition organizers through their websites (see Organizers).