Custom Multi Agent Environment Examples for RLlib

LennartJ · March 2, 2024, 9:09am

Hi everyone, i am new to MARL and RLlib. I want to learn how to build custom environments but i acutally did not find any current multi agent custom environment that actually works and serves as a good tutorial.
I tried this multiagentmaze environment from the book “Learning Ray” but it does not work:

github.com

maxpumperla/learning_ray/blob/main/notebooks/ch_04_rllib.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "780507ca",
   "metadata": {},
   "source": [
    "# Reinforcement Learning with Ray RLlib"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2856c7b0",
   "metadata": {},
   "source": [
    "\n",
    "You can run this notebook directly in\n",
    "[Colab](https://colab.research.google.com/github/maxpumperla/learning_ray/blob/main/notebooks/ch_04_rllib.ipynb).\n",
    "<a target=\"_blank\" href=\"https://colab.research.google.com/github/maxpumperla/learning_ray/blob/main/notebooks/ch_04_rllib.ipynb\">\n",
    "<img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"

This file has been truncated. show original

and i tried the following tutorial:

github.com

anyscale/ray-summit-2022-training/blob/main/ray-rllib/ex_02_create_multiagent_rllib_env.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Notebook 02. Create a custom multi-agent RLlib environment\n",
    "\n",
    "© 2019-2022, Anyscale. All Rights Reserved <br>\n",
    "📖 [Back to Table of Contents](./ex_00_rllib_notebooks_table_of_contents.ipynb) <br>\n",
    "\n",
    "➡️ [Next notebook](./ex_03_train_tune_rllib_model.ipynb) <br>\n",
    "⬅️ [Previous notebook](./ex_01_intro_gym_and_rllib.ipynb) <br>\n",
    "\n",
    "### Learning objectives\n",
    "\n",
    "In this notebook, you will learn how to:\n",
    " * [Code a custom RLlib multi-agent environment](#multi_agent_env)\n",
    " * [Select an algorithm and instantiate a config object using that algorithm's config class](#rllib_algo)\n",
    " * [Train a RL model using a multi-agent capable algorithm from RLlib](#rllib_run)\n",

This file has been truncated. show original

but i did not work either. I guess the main problem is that these tutorials were build for the old gym and respective RLlib versions.
Where can i find good tutorials for newer versions of RLlib?

Thanks!

rusu24edward · April 2, 2024, 7:27pm

You could try abmarl. It provides support for quick, custom multi-agent environment building and integrates with rllib to use its algorithms for training. It’s got full tutorials covering many aspects of MARL and how to use abmarl effectively.

PeterF · September 29, 2024, 10:56am

Could you tell us what the env’s STEP method returns?

Topic		Replies	Views
Example of multi-agent environment RLlib	4	913	September 23, 2022
How can I verify that my custom multi-agent environment meets the requirements to run? RLlib	0	242	May 22, 2023
MultiAgent training Issues RLlib	1	344	April 9, 2024
Can I use RLlib with a custom environment without registering it? RLlib	1	505	February 1, 2023
Question about Environment/Observation construction RLlib	1	376	June 17, 2021

Custom Multi Agent Environment Examples for RLlib

Related topics