Will positional head tracking be activated in the dev kit?

"Bakkar90" wrote:
@jherico is this reply to me ? or is it to the forum in general.

In case it is to me, i am aware of the math required to get the position from acceleration.

Anyways, i have been reading the accelerometer and i am intergrating the data twice but the distance i get it quite small compared to what i have moved.

For example i place the IMU on a chair and pushed it around for 2.5 meters but all i got is 20cm.

So I wrote up a pretty primitive application to try to use the acceleration data to track position.

The output, with my Rift sitting on a shelf, looks like this:

Time 001000           Velocity: -0.014 0.066 -0.035           Position: -0.007 0.033 -0.017 
Time 002000           Velocity: -0.028 0.135 -0.075           Position: -0.028 0.132 -0.071 
Time 003000           Velocity: -0.041 0.205 -0.120           Position: -0.063 0.303 -0.170 
Time 004000           Velocity: -0.051 0.272 -0.154           Position: -0.109 0.541 -0.306 
Time 005000           Velocity: -0.059 0.341 -0.187           Position: -0.164 0.846 -0.476 
Time 006000           Velocity: -0.063 0.409 -0.209           Position: -0.224 1.221 -0.673 
Time 007000           Velocity: -0.071 0.478 -0.226           Position: -0.292 1.668 -0.894 
Time 008000           Velocity: -0.077 0.544 -0.239           Position: -0.366 2.180 -1.127 
Time 009000           Velocity: -0.081 0.613 -0.250           Position: -0.444 2.757 -1.370 
Time 010000           Velocity: -0.080 0.679 -0.255           Position: -0.524 3.397 -1.620 

In 10 seconds, with just the jitter in the acceleration data from a device at rest, drifted by several meters.

"jherico" wrote:

"Bakkar90" wrote:
@jherico is this reply to me ? or is it to the forum in general.

In case it is to me, i am aware of the math required to get the position from acceleration.

Anyways, i have been reading the accelerometer and i am intergrating the data twice but the distance i get it quite small compared to what i have moved.

For example i place the IMU on a chair and pushed it around for 2.5 meters but all i got is 20cm.

So I wrote up a pretty primitive application to try to use the acceleration data to track position.

The output, with my Rift sitting on a shelf, looks like this:

Time 001000           Velocity: -0.014 0.066 -0.035           Position: -0.007 0.033 -0.017 
Time 002000           Velocity: -0.028 0.135 -0.075           Position: -0.028 0.132 -0.071 
Time 003000           Velocity: -0.041 0.205 -0.120           Position: -0.063 0.303 -0.170 
Time 004000           Velocity: -0.051 0.272 -0.154           Position: -0.109 0.541 -0.306 
Time 005000           Velocity: -0.059 0.341 -0.187           Position: -0.164 0.846 -0.476 
Time 006000           Velocity: -0.063 0.409 -0.209           Position: -0.224 1.221 -0.673 
Time 007000           Velocity: -0.071 0.478 -0.226           Position: -0.292 1.668 -0.894 
Time 008000           Velocity: -0.077 0.544 -0.239           Position: -0.366 2.180 -1.127 
Time 009000           Velocity: -0.081 0.613 -0.250           Position: -0.444 2.757 -1.370 
Time 010000           Velocity: -0.080 0.679 -0.255           Position: -0.524 3.397 -1.620 

In 10 seconds, with just the jitter in the acceleration data from a device at rest, drifted by several meters.

You need hysteresis (i.e. simulated friction) to prevent that drift by calibrating the IMU range and centering for each axis and by removing the LSB noise bits. Averaging out the noise bits increases latency, so better to just truncate it and/or force a minimum speed threshold before allowing change. Also, force velocity to zero (thereby freezing the position) when the measured velocity is too low (buried in noise). This is essentially what is done when foot-mounting an IMU. Head tracking requires a fixed posterior (i.e. seated) and relatively accurate skeletal modelling (back and neck), preferrably tuned to the actual wearer through a calibration procedure.

Any drift measured while the Rift sits on a shelf is just noise, so should be truncated below the minimum velocity threshold, whether 10 seconds or 10 hours... And of course, consistent drift means it needs calibration (somewhat temperature sensitive)... Technically, it should recalibrate any time it is detected as stationary.

Very slow actual motion is not allowed, but a user can be trained to use it in a useful manner...

Plus the program used a gravity of 9.78 instead of measuring the real gravity, which is why the vertical drift was larger.

"kojack" wrote:
Plus the program used a gravity of 9.78 instead of measuring the real gravity, which is why the vertical drift was larger.

You can measure the gravity all you like. Wikipedia says that surface gravity in Seattle should actually be 9.811, but in fact if I measure the length of the acceleration vector, again, while the Rift is at rest, you get this output:

Time 001000           Gravity: 9.666           Velocity: -0.018 -0.084 0.025           Position: -0.009 -0.043 0.013 
Time 002000           Gravity: 9.728           Velocity: -0.031 -0.165 0.040           Position: -0.034 -0.168 0.046 
Time 003000           Gravity: 9.812           Velocity: -0.042 -0.247 0.051           Position: -0.070 -0.375 0.092 
Time 004000           Gravity: 9.769           Velocity: -0.050 -0.330 0.061           Position: -0.115 -0.662 0.148 
Time 005000           Gravity: 9.765           Velocity: -0.053 -0.417 0.066           Position: -0.167 -1.036 0.211 
Time 006000           Gravity: 9.760           Velocity: -0.057 -0.503 0.070           Position: -0.222 -1.498 0.280 
Time 007000           Gravity: 9.676           Velocity: -0.056 -0.592 0.074           Position: -0.279 -2.048 0.353 
Time 008000           Gravity: 9.660           Velocity: -0.057 -0.679 0.077           Position: -0.335 -2.683 0.428 
Time 009016           Gravity: 9.699           Velocity: -0.054 -0.768 0.077           Position: -0.390 -3.407 0.506 
Time 010016           Gravity: 9.702           Velocity: -0.051 -0.852 0.076           Position: -0.442 -4.222 0.584 

The rotation sensor is far less subject to jitter.

Rotation: 0.0006 -0.0017 -0.0017 
Accel:    0.0723 0.1820 9.7323 

The Rift is sitting with it's face down, hence the Z acceleration, but I can watch the X acceleration waver between positive and negative values, sometimes almost hitting over 7cm/sec^2. That's pretty dirty data. The rotation sensor's drift is an order of magnitude lower. This is probably why the acceleration data is used for exactly one purpose in the sensor fusion, determining the direction of gravity.