MBRL-Navigation Planner

A model based RL approach for robot navigation.

Table of contents

• General Info
• Technologies
• Setup

General Info

This implemented the MBRL algorithms in paper for robot navigation: Neural Network Dynamics for Model-Based Deep Reinforcement Learning with Model-Free Fine-Tuning, Anusha Nagabandi, Gregory Kahn, Ronald S. Fearing, Sergey Levine

https://arxiv.org/abs/1708.02596

Technologies

Project is created with:

• PyTorch
• Python 3.7
• Gazebo
• ROS Melodic

Setup

In other to start trainning, it is reccommended to have a GPU. As we want to boost the simulation faster with constant rate.

To run this project we need: Dependencies: ros-rl-env

Prequisitive: Turtlebot 3 simulation Navigation stack for turtlebot

Branch:

• refractor: code cleannig
• arws: run on workstation

Rename the package folder to rl_move_base to match with package.xml

You have to rename the PATH names in agent.py to match your real path

run:

1/ the navigation rl environment: turtlebot3 simulation, preferably stage 4

$ roslaunch turtlebot3_simulation turtlebot3_stage_$

2/ run the MBRL implemetation

$ rosrun rl_move_base turtlebot3_mbrl.py