Multi-Agent Reinforcement Learning for Turtlebots

This repo contains ROS2 packages for Multi-Agent Reinforcement Learning (MARL) development on simulated and real Turtlebots. The main motivation behind this repo is to implement multi-agent exploration on turtlebots using RL. This includes a custom sim environment built from scratch and uses slam_toolbox for SLAM. Many of the packages here that handle the integration of simulation, sensing, controls etc. have been adapted from my other repo where I built all of this from scratch to implement my version of Extended Kalman-Filter feature-based SLAM on a real turtlebot: https://github.com/ME495-Navigation/slam-project-GogiPuttar

Package List

This repository consists of several ROS2 packages

General Purpose

• nuturtle_description - 3D models of the nuturtle for simulation and visualization.
• turtlelib - CMake Library with functions for control and SLAM of a differential drive robot

EKF SLAM specific

• nusim - Loads the world consisting of the robot, arena and obstacles, and simulates the robot motion, control and sensors, akin to a real robot.
• nuturtle_control - Handles the control and sensing related to the wheels of the robot
• nuslam - Performs EKF SLAM using the turtlelbot

MARL specific

• multisim - Generates and simulates exploration environments with multiple robots. (TODO)
• multicontrol - Controls multiple robots. (TODO)
• multislam - Uses slam_toolbox for multi-agent centralized SLAM. (TODO)
• multiRL - Uses RL for multi-agent exploration. (TODO)