getting started with building a ROS simulation platform for Deep Reinforcement Learning

时间:2022-08-17 06:07:22

Apparently, this ongoing work is to make a preparation for futural research on Deep Reinforcement Learning. The goal of this work is to build a simulation platform that can insert the Deep Reinforcement Learning algorithms as a robot motion planning or navigation module.

I spent all day to position what I should do in this part of work. With an ambiguous goal and a big picture of the whole project, I am almost lost in the information I encountered in knowing about the related fields and knowledge.

In the end of the day, I hope what is now shaped in my mind is enough close to what my boss need. Below are what I believe to explain my work in the future.

Overview of the project

The project finally hopes to determine the robot's motion trajectory in real time by the method of Deep Reinforcement Learning. Given hundreds of different indoor environments the robot is trained to have the ability to move to a specified target without explicitly programmed navigation and obstacle avoidance algorithms.The following paper

The following paper are some related work that I believe could be useful.

  Target-driven visual navigation in indoor scenes using deep reinforcement learning

Active Object Localization with Deep Reinforcement Learning

Deep Neural Network for Real-Time Autonomous Indoor Navigation

Autonomous Navigation Planning with ROS

And a git book is quite complete in AI:

https://www.gitbook.com/book/leonardoaraujosantos/artificial-inteligence/details

ROS simulation

Gathering or building virtual indoor worlds

...

Replacing some modules in Navigation stack for Deep Reinforcement Learning algorithms

Navigation stack is a commonly used navigation module in ROS platform and SLAM tasks, of which the architecture is shown below. I believe some modules shown in this diagram can be replaced by DRL algorithm which I still need to dig deeper to determine.

getting started with building a ROS simulation platform for Deep Reinforcement Learning

A tutorial might walk through me all the way from building a customised robot to navigating this robot in Gazebo is offered here, which I would follow in the next few days and understand more on the relationship among those modules.

[Tutorial] Getting Starting with Autonomous Robots in ROS via Simulations

...