Mouse vs. AI: Foraging Environment (Unity)

Unity source project for the Mouse vs. AI benchmark — a NeurIPS 2025 competition benchmarking visual robustness and neural alignment in RL agents.

Requirements

Dependency	Version
Unity Editor	2019.4.40f1
ML-Agents package	2.0.1 (com.unity.ml-agents)
Barracuda package	3.0.1 (com.unity.barracuda)

Unity 2019.4 LTS is required. Other versions are not tested and may break the ML-Agents integration.

Setup

1. Install Unity 2019.4.40f1 via Unity Hub.
2. Clone this repository:

   git clone https://github.com/SchneiderMarius/mouse-vs-ai-foraging-unity.git

3. Open the project in Unity Hub: Add → Select the cloned folder.
4. Unity will resolve the package dependencies automatically via the Package Manager (ML-Agents 2.0.1 and Barracuda 3.0.1 are fetched from the Unity registry).

What's included

• Full Unity environment used in the Mouse vs. AI: Robust Foraging Competition
• Agent controller scripts (Assets/Scripts/)
• Training configuration (mouse.yml)
• Python training entry point (train.py, requires mlagents Python package)

Training

Install the ML-Agents Python package:

pip install mlagents==0.27.0

Start training:

mlagents-learn mouse.yml --run-id=my_run

The compiled environment build is required for training — see the competition website for pre-built binaries (Windows / macOS / Linux).

Links

• Competition website: https://robustforaging.github.io
• Benchmark evaluation repo (Track 2 — Neural Alignment): https://github.com/SchneiderMarius/mouse-vs-ai-benchmark

Citation

@inproceedings{schneider2025mousevsai,
  title     = {Mouse vs.\ {AI}: A Benchmark for Visual Robustness and Neural Alignment},
  author    = {Schneider, Marius and Canzano, Joe and Hou, Yuchen and Peng, Jing
               and Smith, Spencer LaVere and Beyeler, Michael},
  booktitle = {NeurIPS --- Evaluations \& Datasets Track},
  year      = {2026},
  url       = {https://robustforaging.github.io}
}

License

MIT