Experiment to determine a method for effortless static calibration of the inertial measurement unit on a small unmanned aerial vehicle

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 11 | Nov 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 740
Experiment to determine a method for effortless static calibration of
the inertial measurement unit on a small unmanned aerial vehicle
David Kofi Oppong, Joshua Ampofo, Anthony Agyei-Agyemang, Eunice Akyereko Adjei,
Kwasi Kete Bofah, Victoria Serwaa-Bonsu Acheamfour
Department of Mechanical Engineering, Kwame Nkrumah University of Science and Technology, Kumasi – Ghana
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper details a series of activities carried
out in an effort to develop a means for static calibration of the
micro-electro-mechanical-systems inertial measurementunit
on a small unmanned aerial vehicle. The study involved
collecting raw accelerationandangularvelocitydatafrom the
inertial measurement unit on an Ardupilot flight controller
and processing the measurementstodeterminethecalibration
parameters. An established calibration procedure provided
benchmarkparametersagainstwhichtheexperimentalresults
were compared. We discovered that it is possible to calibrate
the inertial measurement unit of a small unmanned aerial
vehicle by collecting about a minute’s worth of static data.
Although the values of the calibration parameters differ in
some ways from thoseofthebenchmarkcalibrationalgorithm,
an accurate representation of the true inputs to the system is
obtained. We explore an approach that requires no special
equipment or precise maneuvers in the calibration of these
sensors.
Key Words: Static calibration, inertial measurement unit,
micro-electro-mechanical systems, unmannedaerial vehicle
1. INTRODUCTION
The use of small unmanned aerial vehicles (UAVs) has been
proven in several scenarios, including surveying and
mapping, package delivery, surveillance, precision
agriculture, and recreation [1]. Most of these UAVs come
with micro-electro-mechanical system (MEMS) sensors,
usually accelerometers and gyroscopes, which provide low-
cost, small form-factor sensor solutions. Owing to
manufacturing (and installation) imperfections, errors are
inherent in the output of these sensors [2].Threeofthemost
prominent are misalignment,bias,andscalefactorerror. The
goal of calibration is to eliminate these, and accurately
describe the relationship between the true value of the
measurement and the sensor output.
Standard calibration equipment usually costs significantly
more than the MEMS sensors themselves [3], and an
alternative calibration procedure is often desired. Several
researchers have proposed calibration methods that do not
require expensive equipment. [4] used a set of 18 positions
(6 flat faces and 12 edges of the sensor) to collect
accelerometer data, and 9 positions (coordinate axes
perpendicular and parallel to a hinge) to collect gyroscope
data for calibration. [5] used a similar approach with
accelerometers; with the sensor mounted on a patient, they
collected data and used a thresholdtodetermine quasi-static
states, on which they applied an estimator to determine the
sensitivity and offset of the accelerometer. [6] also avoided
expensive calibration equipment by collecting data with the
inertial measurement unit (IMU) moved by hand and placed
in several static positions. These approaches require the
sensors to be placed in several different configurations in
order to determine the calibration parameters. Sometimes
this is not possible or convenient, especially when the
sensors are rigidly attached to the airframe of a small UAV,
and a method that requires absolutely no movement of the
platform is desirable.
We begin by examining models of these MEMS sensors,
which give us an idea of the calibration parameters to be
determined.
2. SENSOR MODELS
It is necessary to define a sufficientlyaccuraterepresentation
of the sensors if they are to be calibrated properly. [7]
provide the following models.
2.1 Accelerometers
(1)
where,
is the accelerometer output in the sensor axes
frame; is the sensor platform acceleration in
body-fixed coordinates, that is, the desired value;
where is the alignment matrix:
the term in is the rotation of the i-
accelerometer sensitivity axis around the j-th
platform axis.

To simplify the alignment matrix, we define the
sensor and body axes such that their x-axes
coincide, and the body y-axis lies in the plane
defined by the sensor x- and y-axes.
As a result [8], and
.
, the sensitivity (scale
factor) matrix; and b is the sensor bias.
By multiplying and , we obtain the 3 x 3
matrix, , where the
products of the elements of the alignment matrix
have been dropped.
Thus, calibratingtheaccelerometerrequiresthecomputation
of nine parameters: the nonzero elements of F and bias.
2.2 Gyroscopes
(2)
where,
is the gyroscope output in the sensor frame;
is the sensor platform rotational velocity in body-
fixed coordinates, that is, the desired value;
where is the alignment matrix
defined as, , where the
term is the rotation of the i gyroscopesensitivity
axis around the j-th platform axis.
Again, we define the sensor and body axes such that their
x-axes coincide, and the body y-axis lies in the plane
defined by the sensor x- and y-axes.
As a result, and
.
is the sensitivity (scale factor)
matrix; and is the sensor bias. By multiplying and ,
we obtain the 3 x 3 matrix, ,
where the products of the elements of the alignment matrix
have been dropped.
Therefore, calibrating the gyroscope is a problem of
determining the six components of E and the three
components of the bias.
3. EXPERIMENTAL PROCEDURE
The experiment was carried out using an APM 2.8 flight
controller running the Ardupilot multicopter firmware. The
hardware was connected to a computervia theUSBinterface
and telemetry logs of the raw inertial measurement unit
were taken. Mission planner [9], an opensource ground
station software, provided the computer interface to the
controller.
3.1 Benchmark data
Following [10], two sets of benchmark data were collected:
one for the calibration of the accelerometerandtheother for
the gyroscope.
For the accelerometer, each sensor axis was placed parallel
and antiparallel to the gravity vector. Approximately 20
seconds of data was collected in each case.
For the gyroscope, the z-axis was placed parallel and
perpendicular to the gravity vector for 20 seconds each.
Then, each sensitive axis was placed antiparallel to gravity,
and the flight controller rotated anticlockwise about that
particular axis through an angle of 3600 inincrementsof900.
After each 90-degree rotation, the sensor was left idle for
approximately 20 seconds. Finally, the z-axis was again
placed parallel and perpendicular to the gravity vector, and
20 seconds of data collected in each case.
3.2 Test data
Given that our aim was to develop a method for static
calibration of the inertial measurement unit, we collected
about a minute’s worth of accelerometer and gyroscope

data1 from the stationary flight controller with its x-axis
pointing north (with the help of a portable magnetic
compass) and the z-axis parallel to the gravity vector (with
the help of a plumbline).
4. BENCHMARK CALIBRATION PROCEDURE
The calibration parameters were first obtained using the
procedure suggested by [10] on the benchmark data.
4.1 Accelerometer bias
The sensor bias was calculated using,
(3)
where i is the sensor axis (x, y or z), is the acceleration
recorded along the i axis when the sensitive axis, i, is placed
parallel to the gravity vector, and , the acceleration
recorded along the i axis when the sensitive axis, i, is placed
anti-parallel to the gravity vector. Because the data was
collected for about 20 seconds in each case, and
represent the average values.
4.2 Accelerometer coupling matrix
Determining the matrix encapsulating the scale factors and
alignment (coefficient of ap in
(1) involves the computation of two
matrices, U+ and U- as follows:
where the notation representstheaccelerometervalue
along the i axis when the sensor is positioned so that its
sensitive axis, j, is parallel to the gravity vector;
1
The duration of 1 minute was a compromise between the
time required for the sensor output to settle to steady state
and the maximum wait time before flight after power-on.
where is the accelerometer value along the i axis
when the sensor is positioned such that its sensitive axis, j,
is aligned anti-parallel to the gravity vector.
It is also necessary to compute the quantity,
(4)
where, A is the norm of the first column of U+, B is the norm
of column 2 of U+ and C is the norm of the last column of U+,
although U– can be used instead of U+.
The coupling matrix is then equal to .
4.3 Gyroscope bias
Gyroscope bias was modelled as a linear equation in time,
(5)
where, i, is the sensor axis.
Knowing the time elapsed over the collection of benchmark
gyroscope data when the z-axis of the sensor was placed
parallel to the gravity vector allows the constants , ,
and to be determined. and are obtained
using the data obtained and the timeelapsedwhenthez-axis
is perpendicular to the gravity vector.
4.4 Gyroscope coupling matrix
The matrix of scale factors and alignment (coefficient of
in (2) is
obtained by dividing the result of integrating the gyroscope
output by the total angular displacement when it is rotated
through 3600 in 900 increments.
5. PROPOSED CALIBRATION PROCEDURE
It is required that the sensor platform bemountedparallel to
the local horizontal plane (with the help of a plumbline),and
the forward direction towards the north (with the aid of a
portable magnetic compass). The sensors may then be
calibrated.
5.1 Accelerometer
The bias in the sensor output is a DC component; it is the
value of the output when the input is zero. It may be
determined in the following way:
1. The mean is computed for the output of each axis of the
accelerometer

2. The Euclidean norm of the three-component mean
vector is calculated and subtracted from the sensor
output along the z-axis
3. The mean, , is computed for the output of Step 2 (i
represents the sensitive axis: x, y or z)
4. The output of Step 2 is then passed through a high-pass
filter using a cut-off frequency, , adapted
from [5].
5. The mean of the filter output, isdetermined;again, i
represents the specific axis: x, y or z
6. The bias is determined by solving Equation Error!
Reference source not found. for b, using ,
and,
, equal to the 3 x 3 identity matrix.
The remaining parameters (components of matrix F) are
determined next.
7. Because the sensor platform has been positioned
parallel to the local horizontal,thegravityvectorsensed
has only a vertical component, equal to the acceleration
due to gravity [11]; this is taken to be equal to the norm
of the sensor output as calculated in Step 2 above.
8. The first two components of the diagonal of matrix F
are forced to a constant value: their effect is absorbed
by the off-diagonal elements, which are now labelled
, and .
Hence,
9. As a consequence of Step 7, the value of is “don’t
care”, and can be set equal to 0. Thus,
10. Equation Error! Reference source not found. maynow
be solved for the three unknowns using,
, and equal to thebias(calculated
1. The mean, , is computed for the output of eachaxisof
the gyroscope (i represents the specific axis: x, y or z)
2. The sensor output (for each axis) is then passed
through a high-pass filter using a cut-off frequency,
, adapted from [5].
3. The mean of the filter output, is determined
4. The bias is determined by solving Equation Error!
Reference source not found. for b, using ,
and,
, equal to the 3 x 3 identity matrix.
Next, the components of matrix E are determined.
5. The first two components of the diagonal elements of E
are forced to a constant value, and their effect is
absorbed by the off-diagonal elements, which are now
represented as .
Hence,
.
6. With the aid of a GPS receiver, the location (latitude-
longitude) of the sensor platform is determined and
used to compute the direction cosine matrix,
that transforms the Earth’s rotation vector,
fromtheEarthframetothe
local navigational frame. It is realized that the East
component of Earth’s rotation rate turns out to be zero,
and hence, the value of is “don’t care” and can be set
equal to 0. As a result, matrix E is reduced to,
.
7. Equation Error! Reference source not found. maynow
be solved for the three unknowns using,
, and equal to the bias
(calculated earlier).
earlier).
5.2 Gyroscope

6. RESULTS AND DISCUSSION
6.1 Results
The tables below summarize the experimental results.
Table -1: Gyroscope bias: benchmark vs. proposed
calibration procedures
Sensor
axis Benchmark Proposed
%
Error
Benchmark
x 0.1901E-3 0.2E-3 +10.8 0.0016E-3
y -0.1953E-3 -0.4E-3 +108.2 -0.0007E-3
z -0.5823E-3 -1.1E-3 +81.0 -0.0017E-3
Table -2: Gyroscope coupling matrix: benchmark vs.
proposed procedures
Benchmark Proposed % Error
Table -3: Accelerometer bias: benchmark vs. proposed
procedures
Sensor
axis
b (m/s2)
x -0.0060 -0.1015 +1.58E3
y 0.1310 0.0470 -64.1
z 0.8262 0.0007 -99.9
Table -4: Accelerometer coupling matrix: benchmark
versus proposed procedures
6.2 Discussion
The tables compare the results obtained using the
benchmark calibration algorithm with those obtained using
the proposed algorithm. The gyroscope bias was most
accurately computed for the x-axis, with the proposed
algorithm recording a deviation of only 10.8%. Althoughthe
y- and z-axis biases were over-predicted, the signs agreed
with the benchmark values. The signs again agreed with the
accelerometer biases, although the deviation was much
greater. Bias drift of the gyroscopes was observed to be
minimal as its coefficientswereoftheorderof .
The diagonal elements of the accelerometer coupling
matrices agreed for both algorithms with a maximum error
of 3.62%. The off-diagonal elements showed the greatest
Table -5: Inertial sensor datasheet specifications
Sensor Bias
Sensitivity
scale
factor
Cross-axis
sensitivity
Gyroscope ±0.35 rad/s ±0.97 ±0.02
Accelerometer
±0.49 m/s2
(x/y-axis)
±0.78 m/s2
(z-axis)
±0.97 ±0.02
It is observed that the outputs of the proposed calibration
algorithm fall within the limits specified by the datasheet.
7. EVALUATION
To further validate the proposed calibration algorithm,
benchmark data used to calibrate the accelerometer was
applied, and Equations Error!Referencesourcenotfound.
and Error! Reference source not found. solvedforthetrue
inputs. For comparison, the benchmark algorithm was also
used. Figures 1 to 6 show the agreement between the
proposed and benchmark algorithms. In some cases,
particularly with angular velocities, the proposed algorithm
provided a smoother representation of the inputs.
2
An absolute value of 1 is the ideal scale factor value
deviation, reaching up to 100% of the benchmark value, and
this is the case because of fixing those values to zero. For the
gyroscope, the deviations were bigger, although the signs
were preserved.
The first two elements of the diagonals of the coupling
matrix, using the proposed algorithm, are taken to be
constants, and a reasonable starting value is an absolute
value of 12. The value 1 works well for the gyroscope, but
leads to inversion of the x- and y-inputstotheaccelerometer,
hence they are negated for the accelerometer.
A quick referral was made to the sensor datasheets to
validate the outputs of the proposed calibration algorithm.
Error! Reference source not found. [12] provides a
summary of the error limits for the accelerometer and
gyroscope.

Fig -1: X-axis acceleration input - benchmark vs. proposed
procedures
Fig -2: Y-axis acceleration input - benchmark vs. proposed
procedures
Fig -3: Z-axis acceleration input - benchmark vs. proposed
procedures
Fig -4: X-axis angular velocity input - benchmark vs.
proposed procedures
Fig -5: Y-axis angular velocity input - benchmark vs.
proposed procedures
Fig -6: Z-axis angular velocity input - benchmark vs.
proposed procedures

8. CONCLUSION
[1] B. Terwilliger, D. Ison, J. Robbins and D. Vincenzi, Small
Unmanned Aircraft Systems Guide: Exploring Designs,
Operations, Regulations, and Economics, Newcastle:
Aviation Supplies & Academics, Inc., 2017.
[2] A. G. Quinchia, G. Falco , E. Falletti, F. Dovis and C.
Ferrer, "A Comparison between Different Error
Modeling of MEMS Applied to GPS/INS Integrated
Systems," Sensors, pp. 9549-9588, 2013.
[3] I. Skog and P. Handel, "Calibration of a MEMS Inertial
Measurement Unit," Rio de Janeiro, Brazil, 2006.
[4] W. T. Fong, S. K. Ong and A. Y. C. Nee, "Methods for in-
field user calibration of an inertial measurement unit
without external equipment," Measurement Science
and Technology, vol. 19, pp. 1-11, 2008.
[5] J. C. Lötters, J. Schipper, P. H. Veltink, W. Olthuis and P.
Bergveld, "Procedure for in-use calibration of triaxial
accelerometers in medical applications," Sensors and
Actuators A: Physical, vol. Volume 68, no. 1–3, pp. 221-
228, 1998.
[6] D. Tedaldi, A. Pretto and E. Menegatti, "A Robust and
Easy to Implement Method for IMU Calibration without
External Equipments," Manila, 2013.
[7] A. Saxena, G. Gupta, V. Gerasimov and S. Ourselin, "In
use parameter estimation of inertial sensors by
detecting multilevel quasi-static states," Lecture notes
in computer science, vol. 3684, pp. 595-601, 2005.
[8] ArduPilot DevelopmentTeam,"MissionPlannerHome,"
2023. [Online]. Available:
https://ardupilot.org/planner/. [Accessed 7 6 2023].
[9] F. Ferraris, U. Grimaldi and M. Parvis, "Procedure for
effortless in-field calibration of three-axis rate gyros
and accelerometers," Sensors and Materials, vol. 7, no.
5, pp. 311-330, 1995.
[10] B. L. Stevens, F. L. Lewis and E. N. Johnson, Aircraft
Control and Simulation:Dynamics,ControlsDesign,and
Autonomous Systems, 3rd ed., Hoboken, New Jersey:
John Wiley & Sons, Inc., 2016.
[11] B. L. Stevens, F. L. Lewis and E. N. Johnson, Aircraft
Control and Simulation: Dynamics, ControlsDesignand
AutonomousSystems,Hoboken,NewJersey:JohnWiley
& Sons, Inc. , 2016.
[12] InvenSense Inc., MPU-6000 and MPU-6050 Product
Specification Revision 3.4, Sunnyvale, CA : InvenSense
Inc., 2013.
The experiment showed that it is possible to statically
calibrate the inertial measurement unit on a small
unmanned aerial vehicle to a good degree of accuracy
without requiring special equipment. Although the
calibration parameters differ from those obtained using a
different calibration algorithm, their signs are largely the
same, and the proposed method provides a good estimate of
the true inputs to the system.
REFERENCES

Experiment to determine a method for effortless static calibration of the inertial measurement unit on a small unmanned aerial vehicle

More Related Content

Similar to Experiment to determine a method for effortless static calibration of the inertial measurement unit on a small unmanned aerial vehicle (20)

More from IRJET Journal (20)

Recently uploaded (20)

Experiment to determine a method for effortless static calibration of the inertial measurement unit on a small unmanned aerial vehicle