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
113 stars 7 forks source link

what are they ( mobile_base and world tf) #22

Closed CAI23sbP closed 4 months ago

CAI23sbP commented 4 months ago

Hi @zzuxzt @philipdames How are you? Anyway, i have two questions about this code.

  1. As you know, in this code uses "mobile_base" and "world" TF. What are they?. i can't see them in ros_tf_tree. Could you explain them?

  2. When i have launched a training, Odom TF was changing Whenever episode (or scenario was changed). How to fasten it until training is ended? i want to fasten it and to merge with map TF. Screenshot from 2024-02-26 19-29-54

zzuxzt commented 4 months ago
  1. The "mobile_base" and "world" are just the names of the models in the Gazebo world and do not belong to the TF. I just used them to get the robot model state in the Gazebo world through a service provided by the Gazebo simulator. More information about it can be found at Get Model State Example
  2. The changes of the odom TF are because the simulated world/scenario has been reset, which means the robot has restarted at a random position and its odom TF should change. It does not affect the training and the localization is provided by the amcl package.

​

CAI23sbP commented 4 months ago

@zzuxzt Hi , Thanks to your reply. I have another single question. A demo model weight which is called drl_vo.zip, How long it takes to train?. I mean that number of steps are spended to train?

zzuxzt commented 4 months ago

When training from the pre-trained model, it takes about 1M steps.

CAI23sbP commented 4 months ago

thank you!