ARTPS: Autonomous Rover Target Prioritization System

Overview

ARTPS (Autonomous Rover Target Prioritization System) is a hybrid AI system that combines depth estimation, anomaly detection, and learnable curiosity scoring for autonomous exploration of planetary surfaces.

Author: Poyraz BAYDEMİR
Affiliation: Selçuk University
Published: July 25, 2025

Key Features

• Convolutional Autoencoder: Compresses and reconstructs Mars rock images
• DPT_Large Depth Estimation: Monocular depth maps (local weights supported)
• Anomaly Detection: PaDiM, PatchCore, and reconstruction-error scoring
• Curiosity Score: Combines exploitation and exploration signals
• Streamlit UI: TR/EN bilingual web interface at repository root

Installation

Requirements

• Python 3.8+
• CUDA-enabled GPU (recommended)

Setup

git clone https://github.com/Poyqraz/ARTPS.git
cd ARTPS
pip install -r requirements.txt

Local DPT_Large weights (optional, not in git)

Place MiDaS DPT_Large state_dict at raw_models/dpt_large_384.pt (~1.3 GB).
See raw_models/README.md. Requires timm (listed in requirements.txt).
Without this file, depth estimation falls back to a lightweight CNN.

Trained project weights (results/*.pth) and image datasets (mars_images/) are also kept locally and excluded from git.

Usage

Run all commands from the repository root (where app.py lives).

Main application (Streamlit)

streamlit run app.py

Tests and demos

python test_working_autoencoder.py
python demo_artps.py
python scripts/verify_i18n.py

Project Structure

├── app.py                 # Active Streamlit application (root)
├── src/                   # Models, UI (i18n, theme), data utilities
├── assets/                # UI theme assets
├── raw_models/            # Local DPT weights (gitignored; README only in git)
├── results/               # Trained weights and generated outputs (gitignored)
├── scripts/               # Paper figures, benchmarks, i18n verification
├── ARTPS/                 # Legacy duplicate copy (not the active app)
└── README.md

Publications

• ResearchGate: DOI
• Zenodo: Archive
• ArXiv: Preprint on hold

Keywords

Mars rover, Autonomous Exploration, depth estimation, Vision Transformers, planetary surfaces, machine learning, Anomaly Detection, Computer Vision

Citation

@article{baydemir2025artps,
  title={ARTPS: Depth-Enhanced Hybrid Anomaly Detection and Learnable Curiosity Score for Autonomous Rover Target Prioritization},
  author={Baydemir, Poyraz},
  year={2025},
  doi={10.13140/RG.2.2.12215.18088}
}

License

This project is licensed under the MIT License — see LICENSE.