Reinforcement Learning Adversarial Predator-Prey Unity Project

This project implements a Unity ML-Agents adversarial multi-agent environment with a spider (Crawler) predator chasing a worm (Prey). Deployment supports running pre-trained models for inference or headless training using standalone builds.

Prerequisites

• Unity Hub and Editor 2023.2.6f2.
• Python 3.10.12 via Conda (for training).

Interactive Deployment (Unity Editor)

1. Download Unity Hub: https://unity.com/es/download.
2. Install Unity Editor 2023.2.6f2 via Hub.
3. Clone repo: https://github.com/Potzon/Proyecto-RL.
4. Import folder Proyecto-RL-main in Unity Hub > Projects.
5. Load Assets/Scenes/ProyectoV1.unity, press Play.

Headless Deployment (Server/Training)

Use Builds/Build-V1/RLV1 project.exe for no-graphics runs.

Training Setup

Environment

conda create -n mlagents python=3.10.12

conda activate mlagents

pip install torch==2.2.1 --index-url https://download.pytorch.org/whl/cu121

ML-Agents

git clone --branch release_23 https://github.com/Unity-Technologies/ml-agents.git

cd ml-agents

pip install ./mlagents-envs

pip install ./ml-agents

Commands

• Visual training: cd mlagents-learn Config/duoCW-V1.yaml --run-id=test --torch-device=cuda

(Open ProyectoV1 scene first.)

• Headless: mlagents-learn Config/duoCW-V1.yaml --run-id=test --no-graphics --env=Builds/Build-V1/RLV1 project.exe --torch-device=cuda

Structure

Directory	Purpose
Assets	Prefabs, Scripts, Models (.onnx), Scenes (ProyectoV1.unity)
Builds/Build-V1	Standalone exe for headless
Config	YAML configs (SoloCrawler, SoloWorm, duoCW-V1.yaml)
Results	Checkpoints, logs