Add vert offset corrections to VLP16 calib file

[msz-rai]

According to the VLP-16 User Manual (page 54) lidar lasers have vertical offsets in their position. This PR corrects the calibration file.

[JWhitleyWork]

@msz-rai Have you tested this on real hardware?

[JWhitleyWork]

Rebased on ros2 branch to get CI updates.

[JWhitleyWork]

I'm also wondering - are we missing these values for all of the other sensors? I don't see them being populated in any other param files.

[msz-rai]

@JWhitleyWork

Have you tested this on real hardware?

No, I haven't. I don't have physical lidar. I'm developing VLP16 in the simulation.

I'm also wondering - are we missing these values for all of the other sensors? I don't see them being populated in any other param files.

Yes, this is correct, but at the moment, I'm focused on VLP16 only. I am not sure if other models need this offset.

[valgur]

The optical drawing in the VLP-16 manual further reinforces that this correction is indeed warranted:

While the correction would be applied in the correct direction in the driver code, I'm suspicious of the multiplication with cos_vert_angle: https://github.com/ros-drivers/velodyne/blob/master/velodyne_pointcloud/src/lib/rawdata.cc#L796 That does not look correct to me and also the HDL-64E sample code does not do this (although it is included in a commented-out formula for some reason): https://www.termocam.it/pdf/manuale-HDL-64E.pdf#page=34

[valgur]

As for the other sensors - only VLP-16 appears to include this info its manual. However, these values can be estimated from the distance of the focal point and the vertical angle of the beam as vert_offset_correction = focal_distance * tan(-vert_correction).

For example, here are the values for VLP-16 LITE (focal distance = 41.910 mm):	Vertical Angle (deg)	Vertical Correction (mm)	f * tan(-angle)
-15	11.2	11.23
1	-0.7	-0.73
-13	9.7	9.68
3	-2.2	-2.20
-11	8.1	8.15
5	-3.7	-3.67
-9	6.6	6.64
7	-5.1	-5.15
-7	5.1	5.15
9	-6.6	-6.64
-5	3.7	3.67
11	-8.1	-8.15
-3	2.2	2.20
13	-9.7	-9.68
-1	0.7	0.73
15	-11.2	-11.23

However, this does not work as well for VLP-16 Hi-Res:	Vertical Angle (deg)	Vertical Correction (mm)	f * tan(-angle)
-10	7.4	7.39
0.67	-0.9	-0.49
-8.67	6.5	6.39
2	-1.8	-1.46
-7.33	5.5	5.39
3.33	-2.7	-2.44
-6	4.6	4.40
4.67	-3.7	-3.42
-4.67	3.7	3.42
6	-4.6	-4.40
-3.33	2.7	2.44
7.33	-5.5	-5.39
-2	1.8	1.46
8.67	-6.5	-6.39
-0.67	0.9	0.49
10	-7.4	-7.39

Here the placement appears to follow approx f * tan(-angle) * 0.93 + 0.5 * sign(-angle) instead. Still, even the rough estimate should produce a more accurate values than without it.

For VLP-32C and Alpha prime the largest correction values would be around 2 cm. I'm a bit confused and surprised by the manual and VeloView not accounting for this.

[valgur]

Have you tested this on real hardware?

I found this article titled "Analysis and calibration of the VLP-16 for automotive applications", which analyzes the vertical angular accuracy of VLP-16. They appear to be using the simpler sensor model without the vertical offset correction (which is not mentioned in this article, but is in another one by the same author).

They measured the vertical angles of the beams at varying distances relative to the center of the sensor:

Based on that table, the errors of the simpler model vs the "theoretical elevation angle" are:

However, after taking the vertical offset into account the expected angles at given distances agree much better with the measurements:

Here's a notebook containing the above analysis: https://gist.github.com/valgur/217470899f9b74c5901769deb62f10be

Based on this, I would feel quite comfortable with including these additional corrections in the driver.

[msz-rai]

@valgur Thank you for the detailed analysis. It sounds good to me, also.