Various classes in the specs have the "mask" entry undocumented:

• rl/torchrl/data/tensor_specs.py

  Line 1297 in 533d3f9

  class OneHotDiscreteTensorSpec(TensorSpec):

• rl/torchrl/data/tensor_specs.py

  Line 2505 in 533d3f9

  class MultiOneHotDiscreteTensorSpec(OneHotDiscreteTensorSpec):

• rl/torchrl/data/tensor_specs.py

  Line 2880 in 533d3f9

  class DiscreteTensorSpec(TensorSpec):

• rl/torchrl/data/tensor_specs.py

  Line 3304 in 533d3f9

  class MultiDiscreteTensorSpec(DiscreteTensorSpec):

The docstrings should explain that this mask can be set to hide part of the discrete specs to hide part of the domain.
The docstrings should also explain how the mask can be updated (using spec.update_mask).
An example should be provided with the mask. See the tests to understand how the mask is used:

@kurtamohler is it something you could help with?