search

chch9907 / turtlebot3-DDPG-LSTM-PER

multi-turtlebot3 collision avoidance and navigation via DDPG-LSTM with Prioritized Experience Replay on ROS
65 stars 8 forks source link

roslaunch turtlebot3_ddpg_stage_1.launch ERROR #2

Open nfhe opened 2 years ago

nfhe commented 2 years ago

Hello, I ran your code here, and then there was a bug. But I didn't find the problem. I hope you can help me solve it if you have time. Thank you very much.

  1. -roslaunch turtlebot3_gazebo turtlebot3_stage_1.launch

  2. -roslaunch turtlebot3_dqn turtlebot3_ddpg_stage_1.launch

The following error occurred: 2022-08-27 11-25-36屏幕截图

Then I found the error here： 2022-08-27 11-25-56屏幕截图

Then I found that the error displayed was that the camera data could not be accessed, but I received the data using rostatic echo.

Hope to get your reply as soon as possible. Thank you again.

nfhe commented 2 years ago

I want to ask, is the environment used stage1-4? Is the target point fixed? How can I feel that the state is not right when running.

chch9907 commented 2 years ago

Hi, he. For your first problem, I did not encounter this issue when I ran this code. And the problem details you provided might not be enough for me. Hence, what I can advise you to do are : 1) check if you use waffle version's turtlebot3 since only waffle version has camera. 2) check if you obtain camera message correctly, which perhaps refers to chapter 1.5 of my blog https://blog.csdn.net/Cameron_Rin/article/details/117027106 .

For your second problem, which stages you use depends on your launch file, thus you should modify it manually. The target point is not fixed and it will change whenever the environment respawns if I remember correctly. Actually, the setting of target point can be self-defined in spawn.py so you can check it. The state you obtain will not be wrong. However, if you use stage 1 where there is no obstacle, the target position is inherently not related to the camera state, hence it will not successfully finish the task. Other cases can be reasoned by above case, which shows this algorithm can be used for image-style targets rather than position style. On the other hand, we can say this algorithm needs an additional navigation base algorithm.

Hope the above can help you :)

nfhe commented 2 years ago

Thank you very much for your reply. I have solved this problem.

cv_image = ros_numpy.numpify(image_data)

I use this replaces the original expression. I use the original ddpg algorithm to navigate, but the effect is not good.After the training, I can not successfully avoid obstacles. Is it necessary to adjust the Hyperparametric? Or does ddpg have such defects?