search

zw199502 / RL_navigation

deep reinforcement learning for autonomous navigation
29 stars 7 forks source link

execute training #4

Open zyw0319 opened 5 months ago

zyw0319 commented 5 months ago

Hello, I have read your article carefully and carried out the reproduction work. In the process of code reproduction, I met the following problems. I hope you can reply me.

  1. When I trained with low-level DRL, I trained on the A5000 server for 7 days, can I adjust the code to speed up the training?
  2. When I was training with high-level DRL, executing python navigation_DQN_5_goal.py showed the following error; I hope you can spare some time from your busy schedule and give me some advice. Thank you again 高层代码错误
zw199502 commented 5 months ago

Hello, I have read your article carefully and carried out the reproduction work. In the process of code reproduction, I met the following problems. I hope you can reply me.

  1. When I trained with low-level DRL, I trained on the A5000 server for 7 days, can I adjust the code to speed up the training?
  2. When I was training with high-level DRL, executing python navigation_DQN_5_goal.py showed the following error; I hope you can spare some time from your busy schedule and give me some advice. Thank you again 高层代码错误
  1. You can't speed up the training. I don't use Gazebo for training anymore because it is a real-time simulator. Instead, I have built a lidar-integrated simulator, https://github.com/zw199502/LSTM_EGO.
  2. You need to '''mkdir model''' at first.
zyw0319 commented 5 months ago

Thank you very much for your reply, I've created the folder as you suggested: "model/navigation_5_goal" when running the high-level DRL, there is a missing “high_dqn_time_step_1040000.h5”file, when I copy the training results of the low-level DRL “dqn_time_step_1040000.h5”to the 'navigation_5_goal' folder and change the file name to “high_dqn_time_step_1040000.h5 ” , can you give me some advice?

zw199502 commented 5 months ago

Thank you very much for your reply, I've created the folder as you suggested: "model/navigation_5_goal" when running the high-level DRL, there is a missing “high_dqn_time_step_1040000.h5”file, when I copy the training results of the low-level DRL “dqn_time_step_1040000.h5”to the 'navigation_5_goal' folder and change the file name to “high_dqn_time_step_1040000.h5 ” , can you give me some advice?

It is a little strange that you failed in running. I haven't touched this project for a long time and I have to admit that the codes are kind of messy. The 'model' directory should be in the same directory of navigation_DQN_5_goal.py.

zyw0319 commented 5 months ago

How do you visualize the training results of low-level DRL?

zw199502 commented 5 months ago

How do you visualize the training results of low-level DRL?

Sorry, I haven't maintained this project for several years and I don't have enough time to help you more at this moment. You'd better figure it out by yourself. Wish you could well tackle it.

zyw0319 commented 5 months ago

First of all, I am very grateful that you can take the time to reply to me. I have encountered an error now, as shown in the figure below. When I modify the path in the navigation_low.py code, the high-level DRL code can run normally. Can the 1040000.h5 weight generated by the low-level DRL of this project be applied to the high-level DRL? I am very sorry to bother you again. 高层代码错误 ![Uploading 高层代码错误3.png…]()