search

ZJU-Robotics-Lab / model-based-social-navigation

Learning World Transition Model for Socially Aware Robot Navigation
57 stars 11 forks source link

Turtlebots leading to Collision. #6

Open chetanchandc opened 2 years ago

chetanchandc commented 2 years ago

Hello,

When I am trying to launch the (/model_based_version/launch/start.launch), the 4 agents(turtlebot) simulate and direct toward the obstacles in the world(square_obstacle_more.world), followed by collision. The same happens with start_one.launch(model_based_version), start.launch(model_free_version).

Finally, the only difference I observed while launching is, turtlebots don't collide with each other(in most cases), but end up with collision to obstacles in the world.

I assume that I was mistaken somewhere while launching or following the procedure. I hope anyone of you would help me with this.

Thanks! @ @IamWangYunKai

YuxiangCui commented 2 years ago

According to your description, it seems that you have launched the training procedure. In the early stage of RL training, the policy tends to act randomly, which may lead to collisions with other robots all obstacles. If you just want to test the final performance, you should load the test models and tune parameters or file paths in 'main.py' accordingly. And you should also check the terminal outputs to make sure that the models are loaded successfully.

chetanchandc commented 2 years ago

Thanks for the reply. Would you please give me detail on how to tune parameters? I have loaded the test model but ended up with errors. Also, I would appreciate it if you could give me your email address so I can share some results. Thanks! @YuxiangCui

YuxiangCui commented 2 years ago

The parameters in the files are already tuned. Errors may come with incomplete coding packages or other environment problems. You can check the terminal outputs to check for sure.