mehdizouitine / gym_ma_toy Goto Github PK

Hi @yzexeter,
We found a major bug in reward fuction and reset episode.
We fix that in our last commit, We invite you to update gym_ma_toy.

Best regards,
Adil

Reset method dont clear the targets

Hi,

Is it possible to make the running in an animation mode. The environment I am trying only renders one image a time, and I have to close the window to push the experiment forward. I notice that the implemented rendering method is with matplotlib.

I am not sure if this environmet works in windows, as I am trying in windows 10. Perhaps, either the OS or the matplotlib prevents animation.

Do you have a plan to make it work as convensional gym rendering style.

The aim of this project is to make toy environments to allow the user to test its t reinforcement learning algorithms.
We need reference scores so that the user can validate her/his method.

One of the main issues in environments implementation is the lack of documentation.
The idea is to provide exhaustive information allowing the user to use the environment easily.

We cannot say that a diagonal movement is possible if we only know that both vertical and horizontal movement are available.

Here is the following case of error that can lead to the superposition of two agents:

If the diagonal agent A1 is in position (x,y) and (x-1,y) and (x,y+1) are empty then we cannot say that (x-1,y+1) and empty.
We need to check in addition to that that the diagonal box is empty.

The implementation of these environments is an essential part of the project.
Several configurations may be available:

• 1 vs 1 symmetrical
• n vs m symmetrical
• 1 vs 1 asymmetrical
• n vs m asymmetrical