RLLIB support for MultiDiscrete spaces higher than 1D

How severe does this issue affect your experience of using Ray?

  • High: It blocks me to complete my task.

I created a custom environment under Gymnasium and defined a 2D (and beyond) MultiDiscrete action space, which Gymnasium does not throw an error:

self.action_space = spaces.MultiDiscrete(np.array([[self.no_available_options] * 10] * 10),)

But RLLib does not seem to like action spaces with more than 1 dimension:

File “C:[path to python]\Lib\site-packages\ray\rllib\models\catalog.py”, line 452, in get_action_dist
(RolloutWorker pid=80152) return partial(dist_cls, input_lens=action_space.nvec), int(sum(action_space.nvec))
(RolloutWorker pid=80152) TypeError: only length-1 arrays can be converted to Python scalars

I probably can flatten the action space, but having a multi-dimensionl action space will be the most intuitive since each of the values in the 2D array corresponds to real life actions.

Is there any way to get around this? Thanks a lot.