VR-Navigation-Pedestrian

Master's thesis project focused on implementing and expanding a Unity Project Tool made for research purposes on Pedestrian Dynamics using ML-Agents, trained following a Reinforcement Learning approach.

University Bicocca, Milan. Elio Gargiulo - 2025 - 110L.

Installation and Development Environment

Tested Versions

• conda 24.11.3 for virtual environments
• Python 3.10.8 for PedPy
• Python 3.9.6 for ML-Agents
• Unity 2022.3.62f2 downloaded via Unity Hub

Virtual Environment — Base (Windows)

conda create --name tesivenv python=3.9.6
conda activate tesivenv
pip install -r requirements.txt

Virtual Environment — Base (macOS)

conda env create -f requirements_macos.yml
conda activate stage

Virtual Environment — PedPy (if needed)

conda create --name pedpyvenv python=3.10.8
conda activate pedpyvenv
pip install pedpy==1.2.0

Project Structure

Several reorganized folders structure the project, most notably:

• Agents: Contains all scripts and objects related to agents, particularly ML-Agents, including brains and prefabs.
• Artifacts: Contains all scripts and objects related to the artifact system.
• Characters: Includes all avatars and 3D models used to represent virtual agents in the environments, along with animations and animation controllers to properly animate and manage them.
• Maps: Contains all environment-related objects, such as buildings, terrains, and props.
• Scenes: Contains all scenes implemented in the project; this is the core folder housing the actual simulated environments.
• PedPy: Contains scripts and output files used by PedPy to analyze agent behavior.
• Resources and Scripts: Contain generic prefabs, audio, and scripts used throughout the project.
• Oculus, XR, and XRI: Contain the logic and components required for VR compatibility.

Utility

• Repository ML-Agents, used to train and test the models: https://github.com/Zeptogram/ML-Agents-Pedestrian/tree/main