Creating a custom environment

[Yabbanootje]

I want to create my own environment, where I want hazards to be in specific locations. I have created a class that inherits BaseTask, just like the example of GoalLevel0 on the documentation page. I then register this class using the register() function. The next thing I do is make() an environment using the id I gave when registering. However, I get the error that I do not give a config to the task. But I do not know where I should get this config from, especially because I do not need to give a config if I want to use SafetyPointGoal0-v0 in the same way for example. I have the feeling that I am missing some steps, but I cannot find which ones. Could someone help me with this?

[muchvo]

Hello. To assist you more effectively, could you please provide the exact error message you're encountering, along with a minimal reproducible example? This would greatly help us in identifying the root cause of the issue.

In the meantime, you might want to try adding the following line of code to your __init__ method:

self._is_load_static_geoms = False

Please let us know if this helps resolve the issue or if you encounter any further problems. Your feedback is valuable to improve the project.

[muchvo]

I apologize, but the approach you mentioned isn't feasible. Based on your question, I'm assuming you've installed the package using pip, which would make it inconvenient for you to modify the file I pointed out. As an alternative, you can download the source code from GitHub and then install it in editable mode using the command pip install -e . from the root directory of the project. This will allow you to make the suggested modifications more easily.

[Yabbanootje]

I am running my code through docker, so it is a bit more difficult for me to do, but I suppose that is the next issue I should fix.
But is there really no way to make a static environment with your code? My idea is to make a curriculum of tasks, where I want the obstacles to be at the edge of the area and come closer to the goal with each new task. Is there maybe a way to get the seeds of the environments I want so that I can use your environments?
In any case thank you for your answers!

[muchvo]

You can retrieve the seed by accessing the env._seed attribute. For more details, please refer to the code provided here in 157. Furthermore, if you need to have obstacles positioned at the edges, you should adjust the placements, locations, and keepout properties according to your requirements. You can find them in this snippet of code (33-35). Currently, the feature to have obstacles of every new task closer towards the goal is not supported within the same task object due to static configuring.

[Yabbanootje]

Okay that helps! Thank you very much for your assistance!

[muchvo]

Additionally, while a single task object cannot change the area in which random objects are generated, you can achieve this by creating new instances of the task object at different training stages. I've also taken note of your mention of a static environment; if you're referring to a non-randomized task setup, you have the option to disable the randomness. Please refer to this discussion discussion for more details.

[muchvo]

Is there any way to do that without changing your source code?

Someone has suggested a method that allows registering new tasks without modifying the source files.