It is interesting to see that how the two different tracking systems (vision-based and magnetic) could cooperate and compensate to get the better head pose. The landmark tracking results are used to solved for the head pose, and after solution selection and some optimization, the corrected head pose is sent back to compromise with the result from the magnetic head tracker and help the landmark prediction for next frame. We could regard this system as a feedback system, and it will go well if the previous prediction is right, but is there also the chance that the system will go worse and give bad prediction.