MarLo (Multi-Agent Reinforcement Learning in Malmo)¶

MarLO (short for Multi-Agent Reinforcement Learning in Malmo) is a high-level API built on top of Project Malmo for RL research in Minecraft. It was used in the 2018 MarLo Challenge and provides pre-built mission environments with Gym-compatible interfaces.

Relationship to MOSAIC¶

MOSAIC uses MarLo’s mission XML files directly through the MalmoEnv backend. The MarLo Python package is not imported at runtime, but the MarLo repository (3rd_party/environments/marLo) is kept as an upstream reference for mission definitions, launcher utilities, and the MarLo Challenge documentation.

        graph TB
    subgraph MarLo["MarLo (upstream reference)"]
        MarLoPkg["marlo Python package"]
        Missions["Mission XML files<br/>(Attic, MazeRunner, etc.)"]
        Launchers["Launcher utilities"]
    end

    subgraph Malmo["Project Malmo (runtime)"]
        JavaMod["Java Minecraft Mod"]
        MalmoEnv["MalmoEnv TCP Protocol"]
    end

    subgraph MOSAIC["MOSAIC"]
        Adapter["MalmoEnvAdapter"]
        GUI["PyQt6 GUI"]
    end

    MarLoPkg -. "wraps" .-> MalmoEnv
    Missions -- "loaded by" --> Adapter
    Adapter -- "TCP :9000" --> MalmoEnv
    MalmoEnv --> JavaMod

    style MarLoPkg stroke-dasharray: 5 5
    style MarLoPkg fill:#f5f5f5

ID Migration¶

The original MarLo environment IDs have been renamed in MOSAIC:

MarLo ID (legacy)	MalmoEnv ID (current)
`MarLo-MazeRunner-v0`	`MalmoEnv-MazeRunner-v0`
`MarLo-CliffWalking-v0`	`MalmoEnv-CliffWalking-v0`
`MarLo-CatchTheMob-v0`	`MalmoEnv-CatchTheMob-v0`
`MarLo-FindTheGoal-v0`	`MalmoEnv-FindTheGoal-v0`
`MarLo-Attic-v0`	`MalmoEnv-Attic-v0`
`MarLo-DefaultFlatWorld-v0`	`MalmoEnv-DefaultFlatWorld-v0`
`MarLo-DefaultWorld-v0`	`MalmoEnv-DefaultWorld-v0`
`MarLo-Eating-v0`	`MalmoEnv-Eating-v0`
`MarLo-Obstacles-v0`	`MalmoEnv-Obstacles-v0`
`MarLo-TrickyArena-v0`	`MalmoEnv-TrickyArena-v0`
`MarLo-Vertical-v0`	`MalmoEnv-Vertical-v0`

For full environment details (action spaces, movement types, objectives), see Environments Reference.

Mission Previews¶

`MazeRunner`	`CliffWalking`	`CatchTheMob`
`FindTheGoal`	`Attic`	`DefaultFlatWorld`
`DefaultWorld`	`Eating`	`Obstacles`
`TrickyArena`	`Vertical`

References¶

Citation¶

If you use these missions in your research, please cite:

@misc{perez2019marlo,
  title={The Multi-Agent Reinforcement Learning in Malm{\"o} (MARL{\"O}) Competition},
  author={Perez-Liebana, Diego and Hofmann, Katja and Mohanty, Sharada Prasanna and Kuno, Noburu and Kramer, Andre and Devlin, Sam and Gaina, Raluca D.},
  journal={arXiv preprint arXiv:1901.08129},
  year={2019}
}