Hanabi4HMI

Hanabi4HMI is a project to support a user-study where humans have the potential to go up against/cooperate with another user or an Agent (rulebased or otherwise).

The implementation is based on hanabi-learning-environment and this project implements a simple server-wrapper around it.

The project was developed as a cooperation between the Honda Research Institute EU and synyx.

Hanabi

Hanabi is a cooperative card game. Therefore, it is suited for experiments concerning human machine cooperation. Find the instructions and more information on the game here:

• Hanabi Game Instructions
• Hanabi Game Official Webpage

Usage

Install the project and start the server. Then you can access the web-ui in your browser under: http://localhost:8000

Potential starting commands are (obviously not an exhaustive list):

• start hanabi game for two agents playing against each other, with one being of type piers and the other of type rainbow.

  hanabi --player-list piers,rainbow --rainbow-type aiide_all1

• start hanabi game for two human players.

  hanabi --player-list human,human

You can also find more information about potential agents and other options via:

hanabi --help

Resources

For implementation of the game we will use the following repositories:

• hanabi learning environment
  • The hanabi learning environment provides an implementation of the hanabi game
  • It comes with example agents
• hanabi ad hoc learning
  • The hanabi-ad-hoc-learning is an implementation of hanabi which is essentially a fork of hanabi learning environment
  • It provides a large set of implemented rule based agents
  • It provides a framework to have these agents play against each other and shows the achieved results
• pyhanabi
  • pyhanabi is an example project for a web UI implementation for matches between human and machine

Running Container based

You can also run the application in a container.

We provided the necessary files. The workflow would be:

$ podman build -t hanabi .
$ podman run -it \
    -v outputs:/outputs \
    -p 8000:8000 \
        hanabi \
            --record-file /outputs/record.csv

After running that your sever would be accessible under localhost:8000 and your result file would be under ./outputs/record.csv.

License

The project is licensed under the BSD 3-clause license - see the LICENSE file for details.

src/hanabi/agents/external contains code taken from the mentioned hanabi-ad-hoc-learning under (Experiments/Rulebased). The code is licensed under Apache License 2.0. The licenses are also specified in the specific files.