I don't see anything wrong with the calculation. Presumably, one of the bond angles contained within a dihedral definition is going near 180 degrees, but we can't tell from this output. The default initial step of 0.5 au could be very large if concentrated in an angle. As a possible workaround, I suggest trying
INTRAFRAG_STEP_LIMIT = 0.2
You could also try
DYNAMIC_LEVEL = 1
so that the exception should trigger a definition of new internal coordinates if necessary. You could also try
OPT_COORDINATES = CARTESIAN
though this is more of a last resort. I will put this calculation into a test suite for future development.