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
131 stars 9 forks source link

How to use VO #10

Closed GXJll closed 11 months ago

GXJll commented 11 months ago

Hello, after reading your paper, I have developed a great interest in VO. Could you please ask which part of the code reflects speed barriers that I have not found, and how did you use this VO in the PPO algorithm? I hope you can help me answer this question. Thank you very much for your reply

zzuxzt commented 11 months ago

Thank you for your interest in our work. As I worte in the paper, I designed a heading direction reward function using VO to encourage the robot to actively tune its heading direction to avoid moving pedestrians and move toward the goal point. The detailed implementation are in theta_reward.

GXJll commented 11 months ago

thank you for your reply. One more question, is VO input into the neural network as a part of the state?

zzuxzt commented 11 months ago

No, it is only used in the reward function design. The input data are the current pedestrian kinemitic maps, lidar historical map, and the sub-goal point. More details can be found in our paper.