Is it possible to calculate a cross-validated (or otherwise robust) average RDM?

[JAQuent]

I would like to calculate a group-level RDM that is a bit more robust then just averaging all subject-level RDMs. The motivation for this e.g. to use it such a matrix as a model RDM for another set of subjects that do not have the corresponding data.

Let's assume I have a object called rdm_list like the one from the pre-made tutorial (https://rsatoolbox.readthedocs.io/en/stable/demo_fmri_patterns.html):

rdm_list = []
  for sub in subjects:
      print(f'roi {region_name} sub {sub}')

      # load the beta patterns for this subject and collapse the run dimension
      betas = data[f'betas_sub-{sub}_{region_name}']
      patterns = betas.reshape(-1, betas.shape[-1])

      # create an rsatoolbox Dataset with the patterns
      ds = Dataset(
          measurements=patterns,
          descriptors=dict(sub=sub, roi=region_name),
          obs_descriptors=dict(
              run=numpy.repeat(numpy.arange(N_RUNS), len(conditions)),
              condition=numpy.tile(conditions, N_RUNS)
          )
      )

      # for every run, calculate the precision matrix from the residuals
      runwise_prec_matrix = []
      resids = data[f'resids_sub-{sub}_{region_name}']
      for r in range(N_RUNS):
          runwise_prec_matrix.append(
              prec_from_residuals(
                  resids[r, :, :],
                  dof=degrees_of_freedom,
                  method='shrinkage_diag'
              )
          )

      # calculate crossnobis RDMs from the patterns and precision matrices
      rdm_list.append(
          calc_rdm(
              dataset=ds,
              noise=runwise_prec_matrix,
              method='crossnobis',
              descriptor='condition',
              cv_descriptor='run',
          )
      )

      del rdm_list[-1].descriptors['noise'] # saves memory

How could I create a single group-level RDM based on rdm_list? Is it possible to calculate such a thing via CV? I specifically want to calculate based on a list of RDMs and not based on a dataset.

Scrolling through the rsatoolbox reference, I didn't see anything that seem to fit.