Actor-critic multi-objective reinforcement learning for non-linear utility functions

This is the code repository for:

Reymond, M., Hayes, C. F., Steckelmacher, D., Roijers, D. M., & Nowé, A. (2023). Actor-critic multi-objective reinforcement learning for non-linear utility functions. Autonomous Agents and Multi-Agent Systems, 37(2), 23.

• The algorithm implementation is in agents/mocac.py

The code is written in Python and uses Pytorch.

⚠️ This repository uses an old version of Gym, before being part of the Farama foundation. The additional multi-objective environments (in envs/) are based on the old API.

Running the code

This repository contains run-scripts for all experiments of the paper. For example, you can run MOCAC on the Split-environment as follows:

python train_split.py --algo mocac

(the --algo flag can be set to moreinforce|mocac|moac).

The logs are then put in a runs/ folder.

You can plot the results, based on the logs as follows:

python plot_split.py runs/split_env/tunnel_10 moreinforce runs/split_env/tunnel_10/mocac runs split_env/tunnel_10/moac

Where you append, for each algorithm, the folder containing all this algorithm's logs.