interpolate the same bad channels
If the goal is source estimation: do not interpolate bad channels.
You may interpolate them for computing sensor-level group statistics, but mark them as bad for source estimation.
Then I should import this resting state data for each subject alongside the walking data
No need to "import": right-click on the "Link to raw file" > Data covariance > Compute from recordings.
calculate the noise covariance matrix using only a section(about 40 seconds) of the resting state
Compute the noise covariance from longer recordings if you can (up to 10min)
This noise covariance matrix would then be applied to the walking data while calculating the sources.
Right-click on the noise covariance computed for the resting recordings > Copy to other folders.