search

reiniscimurs / DRL-robot-navigation

Deep Reinforcement Learning for mobile robot navigation in ROS Gazebo simulator. Using Twin Delayed Deep Deterministic Policy Gradient (TD3) neural network, a robot learns to navigate to a random goal point in a simulated environment while avoiding obstacles.
MIT License
487 stars 97 forks source link

Gazebo environment #98

Closed zichengS1999 closed 2 months ago

zichengS1999 commented 6 months ago

Is the simulation environment you created yourself or an existing open-source simulation environment?

reiniscimurs commented 6 months ago

Do you mean the world file?

It is a slightly modified version of the simple simulation world found in the paper "Towards Monocular Vision based Obstacle Avoidance through Deep Reinforcement Learning" and is originally available here: https://github.com/xie9187/Monocular-Obstacle-Avoidance

zichengS1999 commented 6 months ago

Thank you for your reply! Is this world file open-source and available for use? Can I use it to test my research direction, and can I cite it in my paper? Additionally, I have encountered an issue during my experiments where the robot in Rviz becomes unsynchronized and randomly teleports when I call the reset() function. Have you encountered a similar situation before?

reiniscimurs commented 6 months ago

You need to check the licenses in this and their repo, but you can use this for research. And yes, you should cite the paper where the world file is taken from.

Not sure what you mean by unsynchronized. Each reset call will randomly teleport the robot to a new location and star executing a new episode. That is how training is intended and done on purpose.

zichengS1999 commented 5 months ago

The issue I encountered is the same as the first issue. It is like the two images below, where the robot's position changes from the first image to the second image after calling reset(). I also changed the robot and used a 2D laser.However, in the Gazebo environment, the robot behaves normally and can successfully complete tasks.

773cee9b7da620a6100f8a176dd727c ff1a65d9bd3e8b7b1c210b28d7c31a0

reiniscimurs commented 5 months ago

Most likely it is an issue with transforms and might just be an issue with visualization in rviz. What is your fixed frame in rviz (under global options)? By default we use odom, but it might be different in your case if you have changed the robot model. You would have to look into your code to see what the proper transforms are.