TS4231V1SpatialTransform fails if lighthouses are at 45 deg. angle

[jonnew]

Describe the Bug

TS4231V1SpatialTransform fails if lighthouses are at 45 deg. angle

To Reproduce

If lighthouses are mounted at e.g. 45 deg. relative to the world's Z axis, the the spatial transform does not produce expected results. This was reported by @ChucklesOnGitHub and we will need to fill in the details on failure mode.

Bonsai Version

2.9.0

OpenEphys.Onix1 Version

0.7.0-rc

Attach your workflow

TODO

[ChucklesOnGitHub]

When calibrated with the lighthouses at 45 degrees, the coordinates of the calibration sites come up ok, but the other coordinates have errors.

In this example, the hs at (14, 10, 0) is read as coordinates (9, 14, 7).

And coordinates are coupled - eg if I move left-right, or top-down, more than one coordinate moves in the graph.

The workflow is the following:

---

Different try (i calibrated again), same issue. Each example is a pair of images is the position of the hs and then the coordinates. The first two examples are calibration sites and match closer than the last two examples which don't match at all. The Z max is 16cm.

[cjsha]

I investigated this a little bit. First, I tilted the lighthouses.

Then I moved the headstage in a straight line between two calibration points to see the error. These were the results:

Sliding the headstage from (0,0,0) to (0,60,0)

There's about +/-1cm error in Z and +/-0.5cm error in X.

Sliding the headstage from (0,0,0) to (150,0,0)

There's about +2cm error in Z and +/-.1cm error in Y. The blip in Y is when I accidentally budged the headstage in the Y direction as I was moving the headstage.

The headstage at (70, 60, 0)

I placed the headstage at about (70, 60, 0), a point which is in my behavioral arena that I'm considering far from any calibration points. z was measured at 4 cm where it should be closer to 0 cm.

Discussion

The calibration, at least along the lines between points of calibration, seems somewhat reasonable. There is some error along the way, but this is likely due to nonlinearities in the lighthouse data that Jon has previously alluded to. The error gets worse as we move away from these"calibration lines" though, i.e. z error is +4cm at (70,60,0).

I'm not sure this has anything to do with the lighthouses being tilted - I suspect this a problem even if the lighthouses are not tilted. When I previously tested the calibration with lighthouses pointing down, the calibration points were closer together, the space that I explored with the headstage after calibration was confined to this smaller space, and this space was centered more directly beneath the lighthouses. I suspect all these things reduced the error in my previous experiments.

Other comments

• Be careful not to be thrown off by error when the headstage is in motion. If the headstage is moving quickly enough, a noticeable error is introduced until the headstage is once again (at least close to) stationary.
• At least X and Y coordinates seemed decently independent.
• The error is not due to any warp in the table as the table is not warped by 2cm along its long edge. That would be quite noticeable.

Conclusion

I think it's worth adding admonition in the corresponding tutorial that describes these errors and how the spatial transform best serves its purpose in a smaller arena positioned directly underneath the lighthouses. A long-term solution might involve adding higher-order polynomials to correct for these errors, probably using regression. If we can more precisely characterize the nature of this nonlinearity (i.e. is it some spherical aberration?), we can more properly correct for it. It might be worth renaming it TS4231V1LinearSpatialTransform or something like that to leave room for other forms of spatial transforms.

But this doesn't explain why Ceci's calibrated so poorly. I will have to move the lighthouses lower to better emulate her setup and see if the problem persists. I will do this in another comment.