search

TempleRAIL / drl_vo_nav

[T-RO 2023] DRL-VO: Learning to Navigate Through Crowded Dynamic Scenes Using Velocity Obstacles
https://doi.org/10.1109/TRO.2023.3257549
GNU General Public License v3.0
140 stars 10 forks source link

Training and implementing an other different robot instead of turtlebot2 #28

Closed tendflow closed 2 months ago

tendflow commented 3 months ago

Hello, thank you very much for uploading the code. I have used your project to train a model. Now I want to test your project on a real robot (not turtlebot2) in gazebo environment and real world. What do I need to do for this? Also, if I want to train a project similar to yours, is the dataset in this work available? I would be grateful if you can answer my question at your free time.

zzuxzt commented 3 months ago

Thank you for your interest in our work. I'm sorry for being late as I was busy with my personal affairs these days.

  1. To deploy our DRL-VO policy to the other robots, you need to make sure the input data structure in the cnn_data_pub.py is correct (e.g., downsampling the lidar scans to 80 points and setting the pedestrian tracking topic to your own). But I will update a new branch for deployment and as a baseline in the next few days when I am less busy.

  2. We are using deep reinforcement learning to train our DRL-VO, which does not require a dataset but requires experience in the interaction between the robot and the environment. If you want to train a new policy, you can directly use our current project and simulator and change the network or other reward settings to train it.

tendflow commented 3 months ago

OK,I got it. Thanks for your help.

tendflow commented 2 months ago

Hello, sorry about bothering you again. If I want to use my simulator, network and reward settings to train a new policy, is your experience in this project still available?

zzuxzt commented 2 months ago

Yes, because the lessons learned from our work are universal and should be applicable in different training simulators.

tendflow commented 2 months ago

OK, thanks for your help.