Use of a Laser Tracker to Actively Coordinate the Motion

of a 3-Meter Industrial Robot to Within 50 Microns

by David Novotny, Joshua Gordon, Jason Coder, Jeff Guerrieri, and
Michael Francis; National Institute of Standards and Technology

T
he Antenna Metrology Lab at the National Institute of Stand­ a wavelength, and even more accurately know actual position during
ards and Technology in Boulder, Colorado, is developing a measurement, are driving forces in accurate coordinate measurement.
robotically controlled antenna facility that uses industrial- Furthermore, the delicacy of equipment in these frequency ranges
grade robotics guided by a metrology-grade laser tracker to perform creates problems for physical contact by conventional (at least to the
antenna calibrations over the frequency range 50 GHz to 500 GHz. antenna measurement community) probing. Our eventual reliance on a
These measurements are performed by scanning a millimeter wave combination and coordination of machine vision, movement methods,
(mm wave) probe antenna about a test antenna over an arc using an and positional measurement is forcing a new method of performing
articulated six degree of freedom (6DoF) robotic arm that has a reach high-frequency antenna alignments and calibrations.
of 3 meters. Such near-field measurements yield important antenna Current mm wave (above 100 GHz) antenna characterization
parameters such as gain, polarization, and radiation pattern. Operation facilities tend to focus on planar scanning techniques.2 Planar scan­
at 500 GHz requires knowledge of the probe antenna position to within ning can give accurate pattern information over an approximate
15–20 mm as well as orientation normal to the scan surface (typically ± 30–35° angle range, is generally easier to align, requires less
within ~ 0.05°) to minimize pointing errors. Ultimately, our ability information about the radiation characteristics of the probe, and
to accurately position and orient the probe antenna relative to the test generally requires less complicated facilities. However, if accurate
antenna, as well as our knowledge of actual position and orientation, information is required beyond the central beam area, antenna effi­
determine the upper frequency limit of our measurement capability, ciency is necessary (as in radiometer characterization), wide-angle
and directly affect the uncertainty of our measurements. polarimetric information is required, or full-pattern characterization
of low gain antennas is needed, then other scan geometries, such as
INTRODUCTION spherical or cylindrical, become more attractive.3
The use of a 6DoF robot for probe positioning and a laser
We are designing a 50–500 GHz, multipurpose, antenna pattern tracker (with optional 6DoF measurement capability) for true
and gain calibration range based on an industrial 6DoF articulated positional measurements can provide the ability to perform mul­
robot with a 3-meter reach, a precision (0.0001° resolution) azimuth tiple scan geometries, correct for positional and pointing errors
rotator, and an alignment hexapod, as seen in figure 1. Antenna testing in the AUT and probe, and determine actual in situ measurement
requirements are increasing in frequency above 110 GHz as high-speed positions to further improve overall result accuracy.4 This coordi­
satellite-to-satellite communications, short-range terrestrial point-to- nated 6DoF movement with independent 6DoF positional meas­
point links, and absolute calibration techniques for remote sensing urements might provide considerable reduction in uncertainty
and climate monitoring research come to the forefront.1 Our need to over conventional antenna characterization facilities that use
accurately position and align antenna under test (AUT) to fractions of stacked positional stages for each type of movement.
26 THE JOURNAL OF THE CMSC/AUTUMN 2013
 obstructions due to the hexapod and cabling limit single-mounting
point CCR scans to approximately 200° scans of the θ rotation. We
can validate that the out-of-plane table movement is acceptable (< 20
mm), and multiple scans show acceptable variations in the pointing
of the z axis (< 0.008°). It is imperative that the rotator measurement
be as accurate as possible. It is the virtual baseline by which all the
other alignments are based. Taking large numbers of points over as
large a physical distance as possible is required to minimize the fitting
errors of the central axis.

Figure 1. Physical (left) and conceptual (right) layout of the Hexapod coordinate system definition
antenna characterization facility setup in spherical meas- The absolute origin of the hexapod coordinate system was deter­
urement mode, as the central axis of the rotator sets the mined by measuring three arcs trajectories of a mounted CCR on
location and orientation of the AUT's z axis and origin of the hexapod plane while the hexapod was commanded through its
the probe scanning; major tasks include aligning hexapod angular range of motion about the three angles, Rx, Ry, and Rz. These
movement to rotator axis for AUT alignment, clocking AUT angles each correspond to movement about the x, y, and z Cartesian
rotation to probe scan plane (AUT polarization), aligning axes, respectively. Full circles were then fitted to these arcs and the
probe rotation to probe scan plane and center (probe polar- normal of these circles determined. From the intersection of the three
ization), and correcting probe location for kinematic offsets circle normals, the origin and orientation of the hexapod frame could
in robot movement (mechanical position correction) be determined. In addition, movement of the hexapod along its three
Cartesian coordinates relative to the x = 0, y = 0, and z = 0 home posi­
SYSTEM OVERVIEW tion verified the proper orientation and direction of the three axes.
With this, the full hexapod frame seen in figure 4 was established,
The task of accurately and repeatably moving a probe antenna which allowed for information about the movement of the hexapod
over a desired path at these frequency ranges is complicated relative to the laser tracker and other frames in the system.
because physical contact with the antennas is problematic, the
limited size of the antennas makes accurate angular alignment AUT aperture coordinate definition and
difficult, and the aperture origin is often virtual, so it must be alignment to the z axis
inferred from measurements, as seen in the graphic at the top of The AUT aperture edges are measured and the centroid of the
this article. The extrapolations of the probe coordinate systems rectangular aperture is found, as seen in figure 5. A working frame
and alignment of the multiple coordinate systems (e.g., rotator, is made at the centroid of the measured AUT aperture. The z axis
hexapod, AUT, robot, probe, and scan geometry) complicates of the AUT must be translated and rotated to be co-linear with the
the desired coordinate positioning and measurement between the central z axis of the rotator. An ideal AUT frame is then erected as
probe and the AUT. a goal frame, and the hexapod is used to guide the AUT. Based on a
To accomplish the desired scanning process of the probe around transformation between the measured AUT frame and the goal frame,
the AUT, the antenna range is built on two major robotics plat­ a 6DoF set of offsets can be calculated by which to translate and rotate
forms. The first section consists of AUT positioning and clocking the hexapod, thereby aligning the AUT to the goal frame.
mechanisms. In figure 1, the rotation stage sits upon an xy stage that First, three widely spaced points on the hexapod plane are measured
nominally places the rotation center and AUT in an optimal posi­ (relative to the AUT aperture to hexapod distance) from which an
tion to get maximum angular coverage from the robot. By fixing the identical set of these three points is generated, but which are moved
position of the xy stage, the rotation stage is fixed, as is the central z according to the AUT-to-goal transform. Two new frames are cre­
axis of the entire measurement. The 6DoF hexapod then aligns the ated from the first set of three points and a second translated set,
AUT to the z axis, both in lateral translation and rotational pointing respectively. The transform between these two new frames is then
alignment. Once aligned to the rotator, the z origin for the AUT is calculated to give the final x, y, z, Rx, Ry, and Rz coordinates to which
established. The robot and probe must now be aligned to the AUT the hexapod must be translated to align the AUT to the z axis of the
reference frame. The resultant series of stacked alignments produces rotator. The result of applying the AUT-to-goal transform can be seen
systematic pointing and position errors that need to be addressed.5 in figure 6.
The three reference points are translated until the calculated
ALIGNMENT AUT centroid is aligned with the central z axis. Errors can accu­
mulate here, as the AUT-to-reference-point distance is large com­
To accurately scan the probe around the AUT and to measure pared to the basis for the hexapod coordinate frame. To minimize
the radiating characteristics of the AUT, alignment of the multiple these translation errors, the aperture needs to be measured again
subsystem components must be performed. to verify the alignment of the aperture after the move.

Main z axis definition from rotator Robot coordinate system definition

measurements Once the AUT is aligned to the rotation stage and the central
The rotator is scanned with a corner cube reflector (CCR) affixed to z axis, robot alignment can be performed. The robot movement
the rotator table (below the hexapod), as seen in figure 3. Propagation needs to be coordinated with the laser tracker. There are two
THE JOURNAL OF THE CMSC/AUTUMN 2013 27
 Component Task(s) Nominal figures of merit
Rotation stage • Sets central z axis of the entire measurement system • Better than 20 mm flatness
• Provides ø rotation of the AUT • 3,600,000 steps/rotation (0.0001°)
• Sets polarization reference of the AUT • Repeatability better than 0.0005°
Hexapod • Provides 6DoF alignment between AUT and • Allows virtual pivot point alignment relative to AUT center
central z axis • <1 mm resolution
• 50 mm/±15° travel
• 30 kg AUT capacity
6DoF robot • Provides scan geometry (θ scan about AUT for spherical, 2D • 30 kg load capacity (includes probe, mm wave
scan plane for planar) for the probe hardware, supports, and cabling
• Aligns scan geometry to z axis • ±70 mm rated repeatability (not accuracy)
• Provides probe polarization • 3 - m reach can provide up to 1.25-m measurement radius at
• Full 6DoF movement allows software correction of actual the probe aperture
measurement position • Up to 1.5 x 2 m planar scan size

Laser tracker • Measures 3D position data • Within the 2–3 m working range, repeatability and
• With three nonlinear CCRs, accurate stationary 6DoF accuracy < 20 mm for direct CCR measurement
can be measured • 6Dof sensor ± 0.01° (18 mm/ 100 mm) pointing error and
• 6DoF sensors can measure in real time at moderates 50 mm positioning error relative to central reflector
speeds (<1 m/s)

Figure 2. Table showing subsystem specifications of the NIST mm wave antenna measurement range

obvious methods of performing the frame translations between measurement the error due to movement is generally acceptable. This
the robot and the laser tracker. We can create a frame similar to limits probe movement to approximately 30 millimeters/second for
the hexapod movement of the AUT. We can also move the robot measurements at 100 GHz.
in its base x, y, and z motions to calculate a virtual laser tracker The robot has the ability to send a pre-trigger prior to arriving at
frame for the base robot movement that can be translated rela­ a given point. A pulse can be sent out prior to arriving at the target
tive to the AUT frame. This will allow the altering of the robot point, to within the process timing cycle accuracy of the robot
geometry to match the scanning needs of the AUT in base robot (~ 3 mS). This allows the positional placement, mm wave signal
coordinates. This allows for less error-prone movements when capture, and measurement of position to occur with minimal
shifting a large robot around small, sensitive, and fragile objects. timing differences. The timing is initially estimated by robot
Additionally, timing between the robot and the laser tracker must speed. For a spherical scan of an AUT, the robot is scanned about
be coordinated. The robot movement to a given position must be the AUT, the AUT is rotated, and then the robot reverses its path.
coordinated with the laser tracker capture of the position data and As seen in figure 7, we adjust the pre-triggering of the robot and
a measurement of the radiated energy between the AUT and probe delay of the instrumentation to ensure the data points are taken at
antennas.6 The robot motion needs to be continuous; if the robot is the same point on the forward and reverse paths.
stopped at each measurement point, the cables that phase-lock the Once the basic timing of the robot is established, the robot is
mm wave measurements continue to vibrate and cause errors. In this measured in its native x, y, and z movement, and a reference robot
scenario, a pause must be inserted at each point (1–2 seconds) to frame is established. This allows the errors in robot position to
allow for the system to stabilize prior to measurement. This can cause be translated directly into correction vectors in robot-coordinate
a 250,000-point scan to have an additional 80 hours of measurement space. As seen in figure 7, the robot repeatability at this speed is
time (the number of points tends to scale as the square of frequency). approximately 30 mm with a few outliers around 100 mm.
We have found that as long as the movement of the probe is less
than 1/50th of the measurement, wavelength during the mm wave Probe coordinate
system definition
One of the most diffi­
cult alignment tasks is the
determination of the probe
location and pointing rela­
tive to the robot and rela­
tive to the AUT. As seen
in the figure at the top of Figure 4. By moving the hexapod
the article, the probe can be in its base coordinate x, y, z, Rx,
Figure 3. Determining the central z axis of the system: a CCR is delicate, and the force of a Ry, and Rz motions, a coordinate
placed on the rotation stage and moved through the maximum CCR being pressed against frame relative to the hexapod can
rotation without blockage to the laser tracker from the hexapod it can cause damage. We are be established; this allows for direct
and cabling; the accurate determination of the vertical z axis developing a combination of calculation of the AUT translation
(right) is critical to the accuracy of the entire measurement, and machine vision techniques needed to align it to the central z
every component is aligned to the rotation stage with a 6DoF offset sensor axis in hexapod coordinates
28 THE JOURNAL OF THE CMSC/AUTUMN 2013
 the table in figure 3, and the offset between the sensor and probe. The
robot process timing accuracy of 3 mS may account for the outliers
around 100 mm. Occasionally (< 0.5%) the robot may take an addi­
tional timing interval to process information; this additional 3 mS at
a 30 mm/s velocity translates to a 90 mm offset. The robot may to be
going to the programmed point much closer than figure 8 suggests for
the 100 mm to 130 mm; there is just a slight variation in the pre-trigger
delay from the robot, causing the tracker and RF equipment to trigger
at a slightly different delay.
Figure 5. Measurement of the initial position of the aperture:
the AUT aperture is approximately 8 mm × 12 mm, and the
Scan geometry alignment to the z axis and
initial position is measured relative to the rotator and hexapod
AUT aperture
frame; the lone point in the left graph is where the central z
The last alignment step prior to the actual measurement is to align
axis of the rotator intersects the plane of the aperture, while the
the scan geometry to the z axis of the rotator and AUT aperture.
frame of the aperture plane is seen as an offset from this ideal
The measurement arc is roughly located over the z-aligned AUT
aperture location
by measuring the initial, uncorrected arc path and calculating the
centroid of the path. The arc is moved, in robot coordinates, to be
to determine probe location and orientation.7 For verification of the centered on the AUT aperture, as seen in figure 9. No corrections
positioning capability of the probe, a CCR is placed at the probe loca­ are initially made for scan tilt relative to the AUT. Once a rough
tion, and a 6DoF sensor is offset approximately 250 millimeters away alignment is done (within ~ 300 mm), a scan is performed in both
from the CCR. The CCR is driven to a given location to within robot directions while capturing the probe position and probe-pointing
step capability (~ 17 mm) from several different directions. The laser orientation. The AUT is then rotated to achieve alignment with the
tracker measures the CCR in addition to the 6DoF sensor location and scan frame using the central rotator, and the ø = 0° coordinate for
orientation. The robot joint set is also measured. The alignment of the measurement is set.
the probe tip relative to the 6DoF sensor and the robot tool interface This is where the 6DoF coordinated movement of the robot
are simultaneously measured. The knowledge of both of these offsets has markedly attractive advantages over current systems. The
allows for first-order positioning and attitude control of the mm wave robot can correct 6DoF position and pointing errors simultane­
probe by the robot, and accurate measurement of the probe location ously. The initial errors are relatively large (on the order of 10–20
and attitude by the 6DoF sensor. millimeters), mainly due to the installation tilt of the robot rela­
To validate the alignment and accuracy of the 6DoF sensor rela­ tive to the rotator. The errors in the positions and attitudes of each
tive to the laser tracker, the probe is scanned in the same arc used for
the timing analysis. At each point in the forward and reverse arc, the
6DoF sensor is measured, and then the arc is repeated while measuring
the CCR. Figure 8 demonstrates the differences between the probe
position as measured by the CCR and 6DoF sensor. The measured
differences are consistent with the manufacturer specifications seen in

Figure 6. Aperture translation and rotation via the hexapod:

the calculation of the frame-to-frame translations between
initial and desired position in the hexapod movement frame
allows for direct hexapod movement relative to its base
coordinates; the translation between the aperture and the Figure 7. Timing between the robot and the laser tracker: by
base plate is done by measuring three nonlinear points on adjusting the pre-trigger delay of robot and the tracker data
the base plate relative to the initial aperture position, and an acquisition delay (left), the relative position error between
iterative translation was done; post-alignment verification of the forward and reverse scans can be minimized; once
the AUT aperture shows that alignment of the AUT to the the tracker delay is set (right) the mm wave measurement is
central z axis is within ± 30 mm and ± 0.04° delayed to align to the laser tracker position capture
THE JOURNAL OF THE CMSC/AUTUMN 2013 29
 individual measurement
point is calculated, and
the corrections (in the
robot frame of movement)
required to bring them to
ideal are determined and
sent back to the robot-
positioning software. The
first iteration of the cor­ Figure 10. The effects of successive levels of position
Figure 8. Measured differences correction: we can see the 90 mm positioning error due to
rections, as seen in figure
between the same scan performed robot timing clearly once the mean error drops below 30
10, demonstrate that the
with an offset 6DoF sensor meas- mm, and we can implement position correction techniques to
measured error in the
uring probe position and a CCR minimize the error of these outliers if they constitute a small
AUT frame is reduced
directly measuring probe position: portion of the total scan area or received power from the
to approximately 85 mm.
the start and stop points of the AUT; this error is independent of the 6DoF offsets in figure 8
Subsequent corrections can
scan cluster is mainly between 40
bring the mean positioning
mm and 50 mm; the few points
above 100 mm may be related
error to ~ 25 mm. We see POSITIONAL UNCERTAINTY ANALYSIS
little improvement after
to timing uncertainty in the robot,
three rounds of positional Figure 12 demonstrates the basic background “noise” we see
as they do not repeat in location
iterations. in positioning. The robot is stopped and there is no power to the
over repeated runs; it should be
The resulting errors can motors, with the breaks on each axis. The resultant variation may
noted that the differences seen
be corrected, to first order, be due to the floor moving, instability in the robot or laser track
here are due to both the 6DoF
using position correction mounts, or the positional noise of the laser tracker itself. This
sensor pointing errors and the
techniques. 8 The time it represents a basic measurement limitation of the current setup.
robot repeatability seen in figure
takes to align the scan The raw position error/uncertainty/repeatability of the robot, as
7; currently, this is one of the larger
geometry is trivial com­ seen in figure 13, can be cor­
sources of positional uncertainty
pared to that for the AUT rected to first order because the
and is limiting the operation of the
to rotator. In less than 30 laser tracker is recording the
system to ~ 250 GHz
minutes, four iterations of actual position of the individual
6DoF position corrections mm wave measurement. Cur­
were completed for the 802-point scan. Once the probe-to-6DoF- rently the position uncertainty
sensor offset is established, the corrections can be done for any between the probe and the AUT
scan geometry. Aligning within the 0.01° uncertainty of the can be estimated by the error in
6DoF sensor provides the ability to have acceptable probe the probe as seen in figures 8 and
pointing errors for the low and moderate gain horns we use for 10. The root-sum-square (RSS)
probes. This is particularly important because pointing errors of the position displacements
cannot be addressed by current processing methods. and the 6DoF sensor uncertainty Figure 11. Resultant meas-
Figure 11 demonstrates the corrected arc aligned to the AUT. is ~ 40 m m. However, better urement points for spherical
The average positioning errors, including 6DoF sensor errors, are estimation of the actual probe scanning of an AUT: the scan
approximately 40 mm. The ø (rotator) axis is clocked to the scan and AUT aperture planes is points are fitted to a planar-
plane, and the 0° point of the robot scan (ø axis) is aligned with needed. Measuring over a small circular arc, and the nominal
the z axis of the rotator. surface and then extending the mean AUT to probe position
results to 100 times the length of errors of ~ ± 40 mm; this
measurement will lead to errors. does not include the system-
The base error of the rotation atic offset of the measured
stage as well as how it affects the probe and AUT frames
alignment of the AUT frame and from the actual position and
robot scan geometry needs to be points of the antennas; timing
assessed. between the robot, positional
tracking systems, and the
measurement have been
FUTURE WORK minimized to limit errors due
Figure 9. The first step of aligning the scan geometry of the The most difficult tasks are to the robot movement during
probe to the AUT: the measured center of the uncorrected defining the AUT aperture plane scanning, and many of the
scan (left) is offset to approximately match the AUT location and the probe aperture plane. measured errors and offset
by moving the entire ensemble by a constant 6DoF offset; this Work is underway to accurately from desired positions can
results in an arc (right) with a positional variation of ~ 200 determine the actual position be mitigated with position
mm from ideal and orientation of the aperture correction
30 THE JOURNAL OF THE CMSC/AUTUMN 2013
of the AUT and the probe; this should generate more accurate
methods of determining the antenna center for location and
antenna edges for determining orientation.7 We will be assessing
the possibility of stepping the robot and scanning the rotator.
However, that puts more stress and strain on the RF cabling
beneath the AUT. In this configuration, however, we can use
three CCRs to align the probe tip, similar to the hexapod-to-AUT
alignment. It may result in better probe aperture determination
(down to the level of the base laser tracker error).

CONCLUSION
We presented a method for controlling the probe position of
a scanning antenna system to within 50 mm relative to a cen­
trally defined antenna coordinate system. It is the first stage in a
complex alignment consisting of multiple stages. Once a basic
Figure 12. Nominal movement of the robot with respect to
scanning surface is defined, positional scanning errors can be
the tracker when the robot is not moving and has brakes
corrected to half a wavelength. We can mathematically correct,
engaged; motion in the room can be seen as outliers in these
to first order, the positioning error of the 6DoF robot (~ 40 mm).
data; the point cloud on the left shows uniformity of the
The ~ 50 mm error in the 6Dof sensor can be reduced by placing
movement with time
the sensor closer to the actual probe tip location. This possibly
results in a reduction of sensor errors to 30 or 40 mm. These
uncertainties in position may allow us to achieve the 1/25th REFERENCES
of a wavelength acceptable error to approximately 300 GHz.
1
The tasks of performing more robust 6DoF sensor-to-probe-tip “Achieving Satellite Instrument Calibration for Climate
alignment and AUT determination are the major hurdles still to Change (ASIC3);” Ohring, G. (ed.); National Oceanic and Atmo­
overcome. spheric Administration; 2007.

THE JOURNAL OF THE CMSC/AUTUMN 2013 31

 ings of the Antenna Measurement Techniques Association; pp.
316–321; 2010.
4
Rousseau, P., Wysock, W.C., and Turano, C.M.; “Using a
Tracking Laser Interferometer to Characterize the Planarity
of a Planar Near-Field Scanner”; 2002 Proceedings of the
Antenna Measurement Techniques Association; pp. 501–506;
2002.
5
Francis, M. and Wittmann, R.; “Uncertainty Analysis for
Spherical Near-Field Measurements”; 2003 Proceedings of
the Antenna Measurement Techniques Association; pp. 43–51,
2003.
6
Novotny, D., Gordon, J., Coder, J., Francis, M., and Guer­
Figure 13. Nominal position spread for a single rieri, J.; “Performance Evaluation of a Robotically Controlled
commanded position on the measurement arc for 71 Millimeter-Wave Near-Field Pattern Range at NIST”; Seventh
sweeps of that point; mean deviation from ideal is ~ 22 mm European Conference on Antennas and Propagation; pp.
with a maximum deviation of ~ 64 mm 4086–4089; 2013.
7
Gordon, J., Novotny, D., Coder, J., Guerrieri, J., and Francis,
M.; “Performance of a Robotically Controlled Near-Field Pattern
2
Slater, D., Hardy, J., Stek, P., Cofield, R., Dengler, R., Jarnot, Range in a Spherical Scan Geometry”; 2013 Antenna Meas-
R., and Swindlehurst, R.; “A Large Aperture 650 GHz Near-Field urement Techniques Association; 2013.
8
Measurement System for the Earth Observing System Microwave Wittman, R., Alpert, B., and Francis, M.; “Planar Near-Field
Limb Sounder;” 2001 Proceedings of the Antenna Measurement Antenna Measurements Using Non-Ideal Measurement Loca­
Techniques Association; pp. 468–473; 2001. tions”; 1996 Proceedings of the Antenna Measurement Tech-
3
Hindman, G., Newell, A., Dicecca, L., and Angevain, J.C.; niques Association; pp. 74–79; 1996.
“Alignment Sensitivity and Correction Methods for Millimeter-
Wave Spherical Near-Field Measurements”; 2010 Proceed- U.S. government work not subject to U.S. copyright

32 THE JOURNAL OF THE CMSC/AUTUMN 2013