4
with an offset of 0.62 m (24.53 in) on the vertical axis,
0.71 m (27.6 in) from the horizontal axis, and 0.52 m
(20.5 in) high.
Dynamic Test Procedures
The main objective of the dynamic tests was to assess the
accuracy of the GNSS receivers while in motion using two
vehicles (UGV and e-truck) depending on their speed,
direction of travel, and GNSS status through RTK or the
satellite-based augmentation system (SBAS). At the begin-
ning of each UGV test, NIOSH researchers cleared the
testing area of unneeded participants, excluding the three
researchers conducting the task. Near the testing area,
researchers maintained the remote connection with the
DAQ throughout the test using a laptop. Inside the test-
ing area, a designated driver of the reduced-scale vehicle
drove the vehicle to the starting position. From that point,
the driver signaled to a spotter to press the record button
on the camera, and the spotter signaled to the researcher
with the laptop to start data collection. Once data collec-
tion had begun, the driver accelerated the vehicle and drove
over each of the ten reflective strips in the forward direc-
tion of travel—from west to east. Data collection ended
after the vehicle had passed the final strip and came to a
stop. The researcher checked the data to make sure that all
ten reflectors had been picked up by the ROLS sensor. The
file was saved, and the test was repeated in reverse or the
return direction—from east to west. The e-truck tests were
conducted like those using the UGV, only with a researcher
physically driving the e-truck. The e-truck tests were con-
ducted at three speeds (5, 10, and 15 mph), while the UGV
tests were conducted at two speeds: low speed (1.27 mph)
and high speed (2.59 mph).
The NIOSH researchers could make more progress to
improve the testing reliability of the ROLS passing over the
center of each reflector and at top speed because during the
dynamic tests, we could not ensure that the ROLS went
over the center of each reflector on every trial. Therefore,
the consistency of our measurement could be more reliable.
In addition, we could also improve the consistency of driv-
ing at consistent speeds because of human error.
DATA ANALYSIS
Static Data Processing and Analysis
To better represent the data collected for the GTI test,
NIOSH researchers subjected the static and the dynamic
data to different data analysis processes. Because the static
tests had less variables that could affect the results, the
processing for the statistical analysis was simpler than the
dynamic tests. For the static tests, the data analysis process
included two steps. First, we performed a coordinate trans-
formation of the GNSS positional data collected and the
surveyed RTS ground truth test points in the same X and
Y coordinates. Next, we calculated the errors in the X coor-
dinates and Y coordinates relative to the ground truth test
points collected using both sets of surveying equipment.
Dynamic Data Processing
After verifying the accuracy of the static measurement
between the RTS and GNSS surveying equipment,
NIOSH researchers processed the dynamic data to estimate
the positional errors of the GNSS receivers relative to the
ground truth points using the following four steps. First,
because the positional data recorded from the GNSS and
the signal from the ROLS were on the same timestamp,
we wrote an algorithm to select the 10 time-events. These
events were characterized as when the ROLS went over a
reflector. Second, using these times, we selected the longi-
tude and latitude associated with each time. Next, we per-
formed a coordinate transformation to convert the latitude
and longitude data from the GNSS receiver into the same
local coordinate system (X and Y) as the ground truth test
points. We then calculated the errors in both the X and Y
directions for each of the 10 reflectors for each trial. Lastly,
the calculated errors were submitted for statistical analysis.
Note that in Figure 3, there were 10 additional standalone
points measured by the GNSS equipment that were used to
assess the dynamic test results.
Statistical Method for Dynamic Data
The objective of this statistical analysis was to create a statis-
tical model that could yield Euclidean error estimates and
95% confidence intervals in instrument measurements for
the dynamic tests conducted using a commercially avail-
able statistical tool. The main variables in this analysis were
the vehicles—e-Truck and UGV. NIOSH researchers per-
formed this analysis using all possible combinations and
levels of the fixed effects. The fixed effects, for example, were
GNSS status such as RTK or SBAS, the different speeds of
the e-truck or UGV, and direction of travel. We summa-
rized the results of the dynamic tests conducted with RTK
and then SBAS for each day. Also, we divided the statistical
analysis in terms of days, GNSS status for that day, and
each speed and direction of travel of the vehicles. We com-
puted the Euclidian error for both vehicles by squaring the
X and Y errors, summing them, and taking the square root
of that sum. We used these errors as the response variable
in our model. A repeated measures mixed model was cre-
ated separately for both vehicles. In the models, we ran the
analysis separately for each date, and both test numbers and
with an offset of 0.62 m (24.53 in) on the vertical axis,
0.71 m (27.6 in) from the horizontal axis, and 0.52 m
(20.5 in) high.
Dynamic Test Procedures
The main objective of the dynamic tests was to assess the
accuracy of the GNSS receivers while in motion using two
vehicles (UGV and e-truck) depending on their speed,
direction of travel, and GNSS status through RTK or the
satellite-based augmentation system (SBAS). At the begin-
ning of each UGV test, NIOSH researchers cleared the
testing area of unneeded participants, excluding the three
researchers conducting the task. Near the testing area,
researchers maintained the remote connection with the
DAQ throughout the test using a laptop. Inside the test-
ing area, a designated driver of the reduced-scale vehicle
drove the vehicle to the starting position. From that point,
the driver signaled to a spotter to press the record button
on the camera, and the spotter signaled to the researcher
with the laptop to start data collection. Once data collec-
tion had begun, the driver accelerated the vehicle and drove
over each of the ten reflective strips in the forward direc-
tion of travel—from west to east. Data collection ended
after the vehicle had passed the final strip and came to a
stop. The researcher checked the data to make sure that all
ten reflectors had been picked up by the ROLS sensor. The
file was saved, and the test was repeated in reverse or the
return direction—from east to west. The e-truck tests were
conducted like those using the UGV, only with a researcher
physically driving the e-truck. The e-truck tests were con-
ducted at three speeds (5, 10, and 15 mph), while the UGV
tests were conducted at two speeds: low speed (1.27 mph)
and high speed (2.59 mph).
The NIOSH researchers could make more progress to
improve the testing reliability of the ROLS passing over the
center of each reflector and at top speed because during the
dynamic tests, we could not ensure that the ROLS went
over the center of each reflector on every trial. Therefore,
the consistency of our measurement could be more reliable.
In addition, we could also improve the consistency of driv-
ing at consistent speeds because of human error.
DATA ANALYSIS
Static Data Processing and Analysis
To better represent the data collected for the GTI test,
NIOSH researchers subjected the static and the dynamic
data to different data analysis processes. Because the static
tests had less variables that could affect the results, the
processing for the statistical analysis was simpler than the
dynamic tests. For the static tests, the data analysis process
included two steps. First, we performed a coordinate trans-
formation of the GNSS positional data collected and the
surveyed RTS ground truth test points in the same X and
Y coordinates. Next, we calculated the errors in the X coor-
dinates and Y coordinates relative to the ground truth test
points collected using both sets of surveying equipment.
Dynamic Data Processing
After verifying the accuracy of the static measurement
between the RTS and GNSS surveying equipment,
NIOSH researchers processed the dynamic data to estimate
the positional errors of the GNSS receivers relative to the
ground truth points using the following four steps. First,
because the positional data recorded from the GNSS and
the signal from the ROLS were on the same timestamp,
we wrote an algorithm to select the 10 time-events. These
events were characterized as when the ROLS went over a
reflector. Second, using these times, we selected the longi-
tude and latitude associated with each time. Next, we per-
formed a coordinate transformation to convert the latitude
and longitude data from the GNSS receiver into the same
local coordinate system (X and Y) as the ground truth test
points. We then calculated the errors in both the X and Y
directions for each of the 10 reflectors for each trial. Lastly,
the calculated errors were submitted for statistical analysis.
Note that in Figure 3, there were 10 additional standalone
points measured by the GNSS equipment that were used to
assess the dynamic test results.
Statistical Method for Dynamic Data
The objective of this statistical analysis was to create a statis-
tical model that could yield Euclidean error estimates and
95% confidence intervals in instrument measurements for
the dynamic tests conducted using a commercially avail-
able statistical tool. The main variables in this analysis were
the vehicles—e-Truck and UGV. NIOSH researchers per-
formed this analysis using all possible combinations and
levels of the fixed effects. The fixed effects, for example, were
GNSS status such as RTK or SBAS, the different speeds of
the e-truck or UGV, and direction of travel. We summa-
rized the results of the dynamic tests conducted with RTK
and then SBAS for each day. Also, we divided the statistical
analysis in terms of days, GNSS status for that day, and
each speed and direction of travel of the vehicles. We com-
puted the Euclidian error for both vehicles by squaring the
X and Y errors, summing them, and taking the square root
of that sum. We used these errors as the response variable
in our model. A repeated measures mixed model was cre-
ated separately for both vehicles. In the models, we ran the
analysis separately for each date, and both test numbers and