Curriculum Learning in torchRL?

Hi, I would like to implement a curriculum learning framework in torchRL. Specifically, I would like to increase the number of agents in an environment after a certain number of epochs or certain average reward reached in a navigation environment like torchRL’s VMASEnv.

Is there a good way to do this within the framework?

Happy to help you make that happen!

Let’s get technical about this: Can you first explain what you mean by increasing the number of agents? Does that mean that you have a dynamic action/observation space?

Also, is there anything you wish you could do but can’t at the moment with the primitives we propose?