SPACESHIP: Synthesizable Parameter Acquisition via Closed-loop Exploration and Self-directed, Hardware-aware Intelligent Protocols for autonomous labs.

SPACESHIP is a flexible and modular framework for autonomous material synthesis. It integrates probabilistic models with hardware-aware experimentation protocols to identify synthesizable regions in high-dimensional parameter spaces — without prior constraints.

Key Features

• Parameter Space Definition: Constructs the experimental parameter space by integrating chemical formulation constraints with hardware-specific capabilities.
• Synthesizable Space Mapping: Identifies and iteratively refines the synthesizable regions through closed-loop experimentation and model-guided prediction.
• Uncertainty-Aware Acquisition: Actively selects informative experiments based on model uncertainty, enabling efficient exploration of under-characterized or high-risk regions.

---

📁 Project Structure

SPACESHIP/
├── ParameterSpace.py                # Defines experimental parameter space (must run first)
├── SynthesizableSpace.py           #  – Main synthesis prediction module
├── BaseModel/                      # Collection of baseline and probabilistic models
│   ├── logistic.py, mlp.py, xgboost.py
│   ├── gpclassifier.py, vgpclassifier.py, emsembleBO.py
│   ├── TabPFN.py, VIME.py
│   └── WideDeep_transfer.py

---

⚙️ Usage Instructions

1. Step 1 – Define parameter constraints:

python ParameterSpace.py

2. Step 2 – Explore synthesizable space:

python SynthesizableSpace.py  

Make sure to run ParameterSpace.py first to initialize or load experimental bounds.

3. Step 3 – Run Autonomous Lab on OCTOPUS:

To execute AUTONOMOUS in a real experimental setup, the code must be run within the OCTOPUS environment with the required hardware connected.

For installation and environment configuration, please refer to the official OCTOPUS repository: https://github.com/KIST-CSRC/Octopus

---

🔧 Installation & Requirements

Python ≥ 3.8 and the following packages are recommended:

pip install torch gpytorch scikit-learn xgboost numpy pandas matplotlib

---

📄 License & Contact

This repository is for academic and research use only.
For questions, please contact:

Nayeon Kim – Korea Institute of Science and Technology / Korea University 📧 Email: kny@kist.re.kr