A Principled Reductionist Approach to Planning in Multi-Objective Markov Decision Processes

This repository contains code for the paper "A Principled Reductionist Approach to Planning in Multi-Objective Markov Decision Processes".

Setup

Python 3.6 or newer is required, as well as the libraries in requirements.txt. To install them, run:

python3 -m pip install -U pip
python3 -m pip install -r requirements.txt

Folder structure

• algo: algorithms for estimating the Pareto front (Benson, NLS, OLS, and Approx)
• env: environments (Deep Sea Treasure, Bonus World, LQR, and Hopper)
• fig: figures and graphs (output from scripts)
• common: things used throughout

Generating the graphs

To generate the graphs in the paper, run:

python3 -m run --figdir fig --seeds 50

This takes about ... . Figures are saved in the subfolder fig.