Calibration Issues

[jedranot]

I also am trying to use the hydra for spatial location and am having similar issues. The attachment below is a test I did to see how much spatial distortion there is as you move away from the hydra. The hydra was located just below this grid and in the middle. The black triangles are the real points and the green circles are the hydra readings. Each point is 10 cm away from each other making the grid 120 cm wide and 110 cm tall. I tried to hold the controller in the same orientation to negate the location of the sensor in the controller.

[TiagoTiago]

was that done with averaging or just a single sample per point?

[TiagoTiago]

Btw, somthing that would look interesting would be if you used the displacements in X, Y and Z in relation to the actual grid, as the RGB components for pixels forming the grid, producing color map of the displacements.

[jedranot]

It was done by taking an average of readings at the point. I also would reject it if the range was more than 50 hydra units.

[andemark]

I have been trying to characterize the physical accuracy also, and am finding similar results. In testing in the XZ plane, there is cross-coupling between reported Z values and physical X coordinates. See attached image. At constant physical Z = 300mm and 450mm, varying physical X results in a non-linear value of reported Z from the sensor. At physical X=200mm, and Z=450mm, Z was about 30mm/6.7% skewed from its true physical value. In the plot, we would expect to see straight vertical lines at 300 and 450mm

I conducted this experiment in a normal computer/desktop environment, and also even further from sources of ferromagnetic interference/distortion with similar results. I also observed these sorts of trends in other dimensions as well.

I think "precision" or sensitivity of 1mm/1deg could possibly be close, but absolute accuracy has very different metrics...

What is the theoretical absolute accuracy we should expect?

[MrMormon]

Suddenly it seems like the Hydra will be useless for musical control...

[Tinus]

If a common pattern is discovered in the distortion, wouldn't it be possible to create a correction function? I mean, if you have a three-dimensional lookup table of reported and actual position and rotation values you could correct any value at runtime with a by interpolating between the nearest values in the table. The graph jedranot posted looks like the values 'fan out' the further they are from the bottom and center lines. Even just quickly correcting for that with an approximate function would help a lot, right? It all depends on wether the character of the distortion is similar for all the devices.

Is this wishful thinking?

[MrMormon]

Hopefully, it is the same or extremely similar for all devices. If not, the manufacturing process has a fundamental flaw.

If so, I would think that one could gather enough samples to interpolate or correct well. But that's just for three-dimensional position. When you throw orientation into the mix, wouldn't we need a freakin' machine to get the six-variable samples in any amount of time with any accuracy?

[TiagoTiago]

Hopefully, it is the same or extremely similar for all devices. If not, the manufacturing process has a fundamental flaw.

...

Unless you're testing several units in the same position in the same environment, there could be factors out of the control of the manufacturers influencing the distortions.

If so, I would think that one could gather enough samples to interpolate or correct well. But that's just for three-dimensional position. When you throw orientation into the mix, wouldn't we need a freakin' machine to get the six-variable samples in any amount of time with any accuracy?

You could fix the positions first, and then figure out the rotation distortions after.

[MrMormon]

Well, yeah, a controlled environment is a must. The rotation distortions aren't even close to independent of the position, though.