In 2019/2020, OpenAI published a paper called "Emergent Tool Use from Multi-Agent Autocurricula", by Baker et al., which generated a lot of momentum in the multi-agent Deep Reinforcement Learning field. This is a project based on OpenAI's multi-agent-emergence-environments (the code released by OpenAI after the publication of the paper) and uses OpenAI's framework to build a new environment (MACL) where developers can deploy their own experiments in order to study good ways to foster cooperation intelligence between agents and machines. There are four experiments already implemented in the MACL.py file. Note that this repository should be used in conjunction with OpenAI gym, the "Spinningup in Deep RL" repository, mujoco-worldgen and multi-agent-emergence-environments.
This work was done during a Semester Project at ETH Zürich and I hope it can serve as additional material to support engineers and researchers in using OpenAI's original code to develop their own projects. For more information, a report is present under the name "Cooperation Learning in multi-agent systems". Besides telling about the structure of the project, the report also has an introductory section on the math and theory behind all the main learning algorithms in the code.
If you are a developer interested in implementing your own multi-agent Deep Reinforcement Learning algorithms, or if you just want to learn more about it, please feel free to use my code as well as my report in supporting your studies/ hobbies/ projects. Just remember to cite them in your references section! Wish you all a good experience, and please don't hesitate to contact me should any questions arise!