Normally this is not an issue when motion is tied to head orientation (e.g. the current Doom 3 BFG demonstration). It doesn't matter what the relation between head and torso orientations are, the user's avatar will always go forward relative to the current view, and the gun is aim point is locked to your head. This can be a problem in some other cases though, especially when the players virtual torso orientation and virtual head orientation are independent, i.e. move-one-way-and-look-another control schemes. Use of these schemes would allow you to use the traditional dual-sick control scheme (left stick torso x-y translation, right stick torso pitch-yaw orientation), with head movement 'on top' of this, e.g. allowing a users avatar to run down a hallway while the user whips their head around to look behind themselves, without interrupting motion.
Use of an absolute position and orientation sensing tracker (magnetic e.g. Raver Hydra, outside-in optical e.g. Vicon, etc) is the optimum solution to this, but these are currently too expensive for most users. One proposed solution is to have a 'yaw reset' button or key, which the user can press to tell the game/software "I am looking straight ahead now". It's an essentially free solution, but has the drawback of reducing immersion and using up a controller button.
I propose what is essentially an automatic yaw-reset button: an narrow horizontal FoV IR sensor (think line-scanner, but just sensing the line as a single pixel) mounted to the underside of the RIFT, and a narrow-horizontal-angle IR LED (line emitter) on a necklace/pendant. The IR sensor will trigger when the user is looking straight ahead, but the narrow emission and viewing angles will prevent it from triggering significantly off-yaw-axis or off-roll-axis, while still allowing pitch motions (sensing at different roll angles could be allowable, but as rolling your head also involves translating your head, this may become tedious to resolve when filtering the inputs). Thus, every time the user looks straight ahead the yaw gyro knows it's absolute orientation and can reset any accumulated drift. Keeping you head held yawed or rolled to one side for a protracted period of time (enough for noticeable gyro error to accumulate) is not a common occurrence (and less common than looking up or down for some time), so this solves the yaw drift problem in most cases.
I've tried to illustrate how the sensor and emitter could be positioned:
To implement this would require power plus a single input pin. I'm not sure if the Hillcrest has a spare GPIO handy (and access to firmware to uncover this functionality), but I can remember Dycus mentioning that the re-purposed HDMI cable the RIFT is currently using has at least one spare conductor pair. This could either be used directly for the sense pin and a downstream ADC, or preferably used to connect a second USB channel. This second channel could either be used for a plug-in module (such as a basic USB ADC for the yaw-reset sensor), or to host an on-HMD USB hub for connecting multiple peripherals.
