🍝 PASTA | Adaptive Smooth Tchebycheff Attention for Multi-Objective Policy Optimization

This repository contains the official PyTorch implementation of the PASTA 🍝 (Policy-optimization via Adaptive Smooth Tchebycheff Attention) algorithm, as presented in the paper:

Alejandro Murillo-González, Mahmoud Ali, and Lantao Liu. Adaptive Smooth Tchebycheff Attention for Multi-Objective Policy Optimization. Robotics: Science and Systems (RSS) 2026

PASTA 🍝 enables stable policy optimization for non-Convex Pareto tradeoffs.

🔥 Quick Start

To quickly run the PASTA 🍝 algorithm and reproduce the included minimal demo run the following commands.

1. Clone the repository:

git clone https://github.com/AlejandroMllo/PASTA.git
cd PASTA

2. Create the Conda environment:

conda env create -f environment.yml

3. Activate the environment:

conda activate pasta-env

4. Run the Multi-Objective demo:

python src/demo_pasta.py

The provided demo illustrates the PASTA 🍝 algorithm within a deliberately minimal multi-objective environment, designed to facilitate rapid evaluation and seamless adaptation.

🚀 What is included

This repository includes:

• src/pasta.py – single-file implementation of the PASTA 🍝 algorithm, including the smoothness controller.
• src/demo_pasta.py – a minimal demo script showing how to run PASTA 🍝 on a simple MORL environment
• README.md – this documentation
• environment.yml – Conda environment file with the required dependencies

📌 Recommended usage

To use the algorithm in your own research or experiments:

1. Inspect demo_pasta.py to see how the environment and training loop are configured.
2. Import the core PASTA 🍝 classes from the src/pasta.py file.
3. Customize the actor and critic architectures, if needed, to handle your environment's observation space.

📚 Citation

If you use this code or the PASTA 🍝 algorithm in your work, please cite:

@INPROCEEDINGS{Murillo-GonzálezA-RSS-26, 
    AUTHOR    = {Alejandro Murillo-González AND Mahmoud Ali AND Lantao Liu}, 
    TITLE     = {{Adaptive Smooth Tchebycheff Attention for Multi-Objective Policy Optimization}}, 
    BOOKTITLE = {Proceedings of Robotics: Science and Systems}, 
    YEAR      = {2026}, 
    ADDRESS   = {Sydney, Australia}, 
    MONTH     = {July}, 
    DOI       = {10.15607/RSS.2026.XXII.134} 
}