[RLlib] Is it possible to change action_space during training?

I have the following action_space:

action_space = gym.spaces.Box(low=1e-12,
                                                         high=1000,
                                                         shape=(1,),
                                                         dtype=np.float32)

So the action space is represented by a vector of size 1 (scalar).
On each step of the episode i want to decrease the high of the action_space by the amount of the action taken.
For example, if action = 100, it is necessary for high to be high=900 after.
Is it possible to design?

Hi @semenov.ss ,

and welcome to the discussion board! I have no clear answer here, but maybe it helps you to consider this problem as a curriculum learning porblem and take a look at the example in RLlib.