This is not straightforward, and the answer depends on what kind of orientation distribution you want to feed into the group tractography.

There are three possibilities:

1) Get average orientation across subjects and do deterministic tractography on the average

2) Get cross-subject distribution of mean orientations and do probabilistic tractography

3) Get cross-subject distribution accounting for intra-individual uncertainty and do probabilistic tractography

[The difference between 2 and 3 is like the difference between fixed and mixed effects in a GLM analysis of group FMRI]

In practice, these 3 options are all doable but we don’t have tools that allow you to perform all the required steps. But here they are in case you are able to implement them yourself.

1) 

- Resample individual dyads onto standard space using vecreg

- Average all the resampled dyads by averaging the rank-1 tensors and getting the principal eigenvector of the average tensor

- Transform to spherical coords using the transform described in https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT/UserGuide?action=AttachFile&do=view&target=fdt_spherical_polars.gif

- Get an average fsamples file across subjects

- Feed to probtrackx2 and use the —noprobinterpol hidden option

2) 

- Resample individual dyads onto standard space using vecreg

- Transform the dyads to spherical coords using the transform described in https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT/UserGuide?action=AttachFile&do=view&target=fdt_spherical_polars.gif

- Concatenate the th,ph files per angle across subjects 

- Create fsamples files by concatenating the mean_fsamples files across subjects

- Feed to probtrackx2

3) 

- Create a random sample of N orientations by selecting subjects and samples at random

- Turn the sample theta/phi into dyads using create_dyadic_vectors

- Resample to standard space with vecreg

- Turn back the samples to spherical coords 

- Concatenate

- Do the same sampling to create fsamples

- Feed to probtrackx2

I hope this helps

Cheers,

Saad