Assume there is only one run of data in your analysis. Some suggestions:
> -polort 3
Cubic fitting might be an overkill. In most cases quadratic fitting is good enough.
> TENT(0, 15, 6)
If 15 seconds of hrf turns out to be too long, it could be shortened to something like TENT(0, 12, 5).
> -stim_file 3 sj01.motion.1D \[0\] -stim_label 3 "Roll" \
> -stim_file 4 sj01.motion.1D \[1\] -stim_label 4 "Pitch" \
> -stim_file 5 sj01.motion.1D \[2\] -stim_label 5 "Yaw" \
> -stim_file 6 sj01.motion.1D \[3\] -stim_label 6 "DS" \
> -stim_file 7 sj01.motion.1D \[4\] -stim_label 7 "DL" \
> -stim_file 8 sj01.motion.1D \[5\] -stim_label 8 "DP" \
It is better to use -stim_base so that those 6 motion parameters would be included as part of the baseline, and the full F value would be an indicator for regressors of interest only.
The 6 files should be inserted like
'sj01.motion.1D[0]'
> -glt 1 ../A_act_vs_con.txt -glt_label 1 'act-con' \
> -glt 1 ../A_act.txt -glt_label 1 'act' \
> -glt 1 ../A_con.txt -glt_label 1 'con' \
Change to
-glt 1 ../A_act_vs_con.txt -glt_label 1 'act-con' \
-glt 1 ../A_act.txt -glt_label 2 'act' \
-glt 1 ../A_con.txt -glt_label 3 'con' \
Gang