Deep Learning With Domain Randomization Gazebo Errors

While starting the above sub-course in my issue title, I have noticed that the Gazebo simulations of the within the workspace don’t seem to be correct. For example, here is Unit 2 with the Jupyter notebook showing the location of the SpamCan in correct position:


Here is a screenshot of how it actually is:


I am not sure that I can go ahead and complete the training if the Gazebo simulation is incorrect. I noticed that in the prior unit, a bunch of the objects have fallen off of the cube, as such:


(As an aside, this is not the third time that I have reported bugs and inconsistencies within the forums and have not had any responses as to if or when they will be fixed. I am starting to get frustrated at using this service to educate myself with ROS.)

These objects are reseted into the correct table zone when you call the service:

rosservice call /dynamic_world_service “{}”

This is called on each training episode randomize the location of the objects.

So there is no issue on gazebo simulation, it simply starts with the objects scattered on the world but as soon as you start and call this service in the training everything will be where it should:

Screenshot from 2020-07-02 11-11-12