0% found this document useful (0 votes)
24 views13 pages

Positioning and Navigation Using IMUs and Low-Cost

This article discusses a novel approach to enhancing navigation systems using multiple Inertial Measurement Units (IMUs) and low-cost sensors to maintain accuracy during GPS denial. It demonstrates that a reliable navigation solution can be achieved for up to 20 minutes at highway speeds by utilizing odometry and magnetometry data alongside IMUs, while addressing challenges such as bias stability and data noise. The paper emphasizes the potential for low-cost consumer navigation systems that operate independently of external inputs, improving navigation in challenging environments.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
24 views13 pages

Positioning and Navigation Using IMUs and Low-Cost

This article discusses a novel approach to enhancing navigation systems using multiple Inertial Measurement Units (IMUs) and low-cost sensors to maintain accuracy during GPS denial. It demonstrates that a reliable navigation solution can be achieved for up to 20 minutes at highway speeds by utilizing odometry and magnetometry data alongside IMUs, while addressing challenges such as bias stability and data noise. The paper emphasizes the potential for low-cost consumer navigation systems that operate independently of external inputs, improving navigation in challenging environments.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 13

This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 1

Positioning and Navigation using IMUs and


Low-Cost Sensors
Patrick Grates, Member, IEEE

Abstract - It is possible to supplement consumer navigation systems that are based solely on Global Navigation Satellite System
(GNSS) with inertial or magnetic field-based sensors so that an accurate navigation solution can be reached during periods of
Global Positioning System (GPS) denial. A fresh approach uses multiple Inertial Measurement Units (IMUs), three spinning and
one unspun, as well as navigation aids for a comprehensive navigation solution. Odometry and magnetometry data is readily
available in two thirds of vehicles manufactured after 2018, and this data may be used in conjunction with independent sensors,
such as and Bluetooth Low Energy (BLE) capable digital compasses. IMUs must be rotated in a controlled fashion and filtered to
account for bias and data noise. Frequent calibration is required to manage bias stability. This paper demonstrates that a reasonable
navigation solution can be arrived at during periods of GPS denial of up to 20 minutes at highway speeds using multiple IMUs and
supplementary sensors.

Index Terms— IMU, Navigation, Odometry, Position, Timing


and were considerably smaller and cheaper than the gimballed
stable platform INSs. These “strap down” INSs were
I. INTRODUCTION1 primarily employed by tactical missiles where they remain

I nertial Navigation Systems (INSs) started with rocket today in various forms. The instruments (gyroscopes,
flight in the mid-1940s, and went into significant accelerometers, and magnetometers) are packaged into a
maritime service with the first navigation over the North sensor called an IMU, and it’s the user’s option to determine
Pole by the USS Nautilus, August 13, 1958 [1]. These INSs whether stable platform, strapdown, or other concepts are best
depended on gimbaled stable platforms with each gimbal for their application. The methods chosen are referred to as
maintained with an angular rate sensing spinning mass mechanization. The IMU, combined with its mechanization,
gyroscope paired with a torquer motor/encoder combination other sensors, computing power, memory, and user interface
in a high frequency response rate feedback loop. Three become the INS. The state-of-the-art IMU is based on Fiber
precision accelerometers were mounted on the stable platform Optic Gyros (FOGs), Ring Laser Gyros (RLGs), or Micro-
in an orthogonal fashion in order to obtain the data required Electronic Mechanical Systems (MEMS) [3]. MEMS based
for Dead Reckoning (DR) navigation. If accurate speed and IMUs are lower in cost. Common MEMS based IMUs are less
direction are obtained, DR is used to plot a course, based on a than $100, and the most sophisticated MEMS based IMUs,
starting point, and combined with periodic course checks with low noise and excellent bias stability, seldom cost more
based on the stars or navigation aids, very accurate navigation than $2000. So there is a large incentive to move from
is possible. Modern stable platform navigators such as expensive FOG or RLG based IMUs to the low-cost MEMS
Shipboard Inertial Navigation System [2], and its based IMUs by developing methods to deal with the noise,
replacement, Electrostatically Supported Gyro Navigator bias, bias stability and mounting problems associated with the
(ESGN), can navigate accurately for long periods of time MEMS based IMU. Rotating or tumbling an IMU has been
without a course check (sometimes referred to as a fix). When previously shown to significantly reduce errors in the
computing power was revolutionized in the late 1970s and accelerometers, and combined with data filtration techniques,
early 1980s, the gimballed stable platform was replaced by a Kalman based or otherwise, make angular rates reported from
virtual platform maintained by a microprocessor so that the the gyros more useable too. Reference [4] initially explored
accelerometers could be “strapped down” to the body they rotation with a stable platform navigator. Two-dimensional
were mounted on, and the computer kept track of where the low-rate rotations with reversals (tumbling) were investigated
virtual platform should be with respect to where it started. in reference [5] to address noise in FOG based stable platform
Long term navigation accuracy suffered, but for short implementation. References [6] and [7] both addressed
durations, “strap down” INSs performed well, met the need, implementing MEMS based IMU rotation combined with an

1
Manuscript received May 14, 2024.
The author is with Arizona State University as a PhD student in Systems
Engineering, 7001 E. Williams Field Rd., Mesa, AZ 85212, USA (e-mail:
[email protected])

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

Extended Kalman Filter (EKF). Finally, Optimal sampling reference [15]. Solid accuracy on rotation phase angle is
rates were investigated in reference [8], and optimal rotation achieved, and this provides an improved navigation solution.
rates in reference [9]. This paper focuses on methods of The problem with filtering accelerometer data is that some
managing noise and bias stability by using multiple MEMS useable data will be lost, and some unwanted data will be
based IMU sensors building on those rotation concepts and retained. Most of the time, the negative aspect of filtering is
filtration techniques successfully demonstrated. by far offset by the positive. Adding rotations makes it worse
The use of rotation to improve DR navigation works well due to errors in sorting out rotation acceleration from sensed
as previously shown for wheel mounts in references [6] and acceleration. For acceleration due to changes in speed, it’s less
[7], and in the lab in references [8] and [9]. The vehicle wheel of a problem, but for acceleration due to changes in direction,
mount IMU advanced the MEMS based IMU DR navigation EKF can make things worse. Fortunately, for modern freeway
concept considerably both in its mechanization techniques as driving, speed is somewhat constant in light traffic.
well as its mechanical simplicity with no rotation motors or Acceleration data processes well for speed. Direction remains
resolvers required [6], [8]. This works well for low vehicle a problem. Extensive experimentation for this paper found it
speeds under known conditions [10]; however, under normal better to segment off, essentially parsing the acceleration data,
driving conditions on open roads, shock will become a portions of the trip based on constant direction and using a
problem. The vehicle wheel itself is the most susceptible to moving average filter of 20 datum performed better than an
shock. Shock can cause significant change in accelerometer EKF. The data could be parsed using logic conditions since
bias particularly at higher vehicle speeds, and bias instability direction changes coincide with large accelerations and match
and noise are the largest contributors to the error budget for with changes in gyroscope angular rates. This improves
the MEMS based IMU. Shock has been a considerable results for both speed and direction. Angular data from the
problem going back to the beginning of INSs. ESGN was gyroscopes behaves much differently. After IMU rotation is
mounted in a significant Shock Isolation System (SIS) [11]. added, there’s no accurate way of removing the spin over
The workhorse navigation system for the U.S. Navy, (relatively) long periods of time (20-30 minutes) without
AN/WSN-7, has a significant SIS [12]. Shock will not be adding error. This eliminates a significant source of data. In
addressed in this paper other than the need for shock isolation. this paper, a fourth IMU is carried without rotation. Although
The INS used in this paper has some shock absorbing the same problems noise, bias stability and error exist, new
properties with a rubber pad on its base. It is operated in the methods of using an Unscented Kalman Filter (UKF) [16],
cab of the vehicle eliminating most shock. Operation in the that employ a series function better suited for the data [17],
cab also eliminates the need to translate between the wheel show better results than an EKF. Reference [16] also suggests
frame and the body frame. using a Smooth Variable Structure Filter (SVSF) after running
Temperature is another factor influencing the stable an UKF. Results based on the UKF/SVSF combination have
performance of an INS. All high-end navigation systems not resulted in useful body attitude data based on the
operate in a temperature-controlled environment as detailed gyroscope data, but optimization of gyroscope data through
in references [12] and [13]. For FOG based navigation multiple IMUs without rotation prior to filtering may provide
systems, atmospheric pressure is an additional constraint [14]. better results.
For MEMS based IMUs, however, pressure has no impact, Navigation systems generally use any useful data that is
and the only temperature impact would be temperature available. In addition to the IMU, these are often referred to
changes between calibration and usage that causes bias to as navigation aids. Long-term navigators, that don’t require
change. Recalibration would be required. This paper will not GPS, such as ESGN depend on other sensors in its system. A
address temperature swings for MEMS based IMUs, but sonar array provides accurate velocity data. When Schuler
acknowledges that a consumer-focused low-cost INS, oscillations occur, ESGN depends on An Electro-Mechanical
primarily using MEMS IMUs, would need to address Log (that monitors the speed the water is passing the ship’s
temperature swings. hull) to dampen other inputs when Schuler oscillations occur.
Data filtering is essential to any tracking systems Some concepts that employ MEMS based IMUs treat
including navigation, missile guidance, and radar. It becomes odometry as a competing method [6]; whereas, it should be a
more important in navigation when adding rotation to the complimentary method of odometry with UKF as used in
IMU to reduce error. It is possible to add error to the system reference [18], UKF with supplementary data as in reference
due to the rotation itself. In order to remove or ignore data [19], and accelerometer/gyroscope pairs with adaptive signal
that’s caused by the rotation, the rotation phase angle must be processing as in reference [20]. Navigation through use of
tracked precisely. Mounting the IMU on one of the “non- odometry combined with IMU data has been successful [6],
turning” wheels of the vehicle is not precise in itself, and leads [18]. In modern vehicles, modern odometry data is readily
to errors. The wheel could also have alignment problems available for post processing, and although not available real-
called camber and toe that would lead to IMU data errors. time, with modern emphasis on autonomy, most vehicle real-
References [7] and [8] manage this problem using an EKF that time data should be available in the future. Odometry remains
provides useable rotation phase angle data. Reference [6] relevant for personal vehicles such as Wheelchairs, Riding
takes it a step further to use an EKF to fuse accelerometer and shopping carts, and motorized scooters. When smart cell
gyroscope angle rates, similar to the methods described in phone signals are spotty, Assisted-GPS (A-GPS) can be

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 3

turned off to continue accurate navigation, but solo GPS can simple moving average filters as appropriate to improve data
take up to ten minutes to synch. Satellite data rates are also accuracy. Kalman filters may be used for data fusion or data
slow, and position updates may lag needed timeliness. filtering techniques. Simple moving average filters are
Odometry data combined with magnetometry (compass) data sometimes used to clean accelerometer data.
can fill this need when other methods are delayed. Indoors, Rotating IMU: When rotating an IMU, the established
when magnetometry data may not be available, odometry data inertial sensor frame (s) axes are aligned with the sensors in
combined with Simultaneous Localization and Mapping the IMU, and the origin is defined as the center of the IMU.
(SLAM) can supplement the navigation solution. Pedometers, The inertial frame (i) is defined to be stationary in space (or
if calibrated frequently for both stride and gait, can work well it can be considered to be moving with constant velocity with
with magnetometry and SLAM for walking navigation with no acceleration). In this case, the inertial frame is stationary
accuracy similar to GPS. This paper will use odometry in space. All inertial sensors produce measurements relative
to an inertial frame resolved along the instrument’s axis of
combined with magnetometry data, odometry combined with
measurement. The body frame (b) is the frame relative to the
MEMS based IMU acceleration data, and IMU acceleration
body that the IMU (or sensor) is mounted on. The IMU may
data alone to navigate and compare results. or may not be rotating on its body frame mount. In this case,
It should be noted that using local frame of reference such the sensor will always be rotating in one axis. The capital
as North East Down (NED) or Earth Center Earth Fixed letter C will refer to a transformation matrix [22], in Euler
(ECEF) precludes the need to account for earth rate, and 20- angles, between reference frames. Equation (1) represents
to-30-minute navigation times eliminate the need to account rotations about the x-axis, (2) about the y-axis, and (3) about
for Schuler oscillations. the z-axis where q=wt. Each matrix is orthogonal so the
There’s been no new navigation product for the consumer CT=C-1.
since GPS became available to the public. Navigating in
mountains, shielded buildings, among tall structures, and in 1 0 0
underground subways remains a navigation problem. 𝐶! (𝜔𝜏) = '0 cos 𝜔𝜏 sin 𝜔𝜏 0 (1)
Combining newer technology now available with existing 0 − sin 𝜔𝜏 cos 𝜔𝜏
technologies, sometimes already available in vehicles, short
term autonomous navigation is feasible, accurate, and cos 𝜔𝜏 0 − sin 𝜔𝜏
𝐶" (𝜔𝜏) = ' 0 1 0 0 (2)
dependable at an affordable price. Wheel chairs and other
personal vehicles also offer a platform for this type of sin 𝜔𝜏 0 cos 𝜔𝜏
system. Navigation inside large buildings, like hospitals or cos 𝜔𝜏 sin 𝜔𝜏 0
convention halls, would cease to be an intimidating problem. 𝐶# (𝜔𝜏) = '− sin 𝜔𝜏 cos 𝜔𝜏 00 (3)
In summary: There is a need, at the low-cost commercial 0 0 1
consumer level, for a navigator to supplement current
personal systems that requires no external inputs, and can Specific force and angular rate in the body frame with respect
provide accurate location and direction information during to the inertial frame and resolved in the body frame are
short periods (20 to 30 minutes) of GPS denial using multiple designated 𝑓$%% and 𝜔$% %
in equation (4) and (5). The
IMUs and supplementary sensors. This innovative product is transformation matrix between from sensor to body is 𝐶'% .
feasible now using combination of existing error
management systems and technologies used separately in the 𝑓$%% = 𝐶'% (𝑓$'' + 𝑓'%
'
) (4)
past.
% ' '
𝜔$% = 𝐶'% (𝜔$' + 𝜔'% ) (5)
II. METHODS AND PROCEDURES
Where 𝑓$'' is in the sensor frame with respect to the inertial
Attempts have been made to use a MEMS IMU with '
frame and resolved in the sensor frame, and where 𝑓'% is in
rotation in a single axis to reduce errors resulting from bias,
scale factor, and misalignment. Traditionally, this would be the body frame with respect to the sensor frame and resolved
considered a strap down system with rotation. Two sources in the sensor frame. The relationships are the same for
[6] and [8], have completed experimentation with this angular rates. Noise for the sensor is designated as the letter
concept successfully using MEMS based IMUs with rotation n, where ns represent the random constants of the gyros and
around one axis. Other single axis rotation experiments are accelerometers in the sensor frame. Bias errors for both the
currently in work, but few have considered simultaneous gyro and accelerometer are designated bs.
rotations using multiple MEMS based IMUs [21]. In
addition, none have combined multiple IMUs with rotation 𝑛!'
with data filtering. An EKF was used in IMU rotation in 𝑛 = 5𝑛"' 6
'
(6)
vehicle wheels [8] to track rotation phase, and EKF was used 𝑛#'
in [6] for data fusion. None have used data filtered from
multiple rotating IMUs combined with odometry, 𝑏!'
magnetometry, and mapping aids to form a comprehensive 𝑏 = 5𝑏"' 6
'
(7)
solution. The technique used here applies Kalman filters or 𝑏#'

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

Error matrices for scale factor and alignment errors used in a 𝑇𝑛!*
space stabilized terrestrial navigator, that is referenced 𝑛* = ' 0 0 (14)
locally, are the same for a MEMS IMU package as described 0
in reference [22]. Equations (8) and (9) represent scale factor
errors, and (10) and (11) alignment errors for gyro and 𝑇𝑏!*
accelerometer respectively. 𝑏* = ' 0 0 (15)
0
𝐾(! 0 0
IMU rotation will reduce errors in two axes as shown in
𝑆( = 5 0 𝐾(" 0 6 (8)
equations (14) and (15), errors will still propagate in the x-
0 0 𝐾(# direction (if rotating about x), the same holds true of the other
two axes. If rotating about y, errors still propagate in the y
𝐾)! 0 0 direction. If rotating about z, then errors propagate in the z
𝑆) = 5 0 𝐾)" 0 6 (9) direction but are reduced through rotation in the other two
0 0 𝐾)# directions. In a locally referenced NED, or East North Up
(ENU) systems, rotation is usually chosen about the z-axis
0 𝐾(!" 𝐾(!# that is up or down since accuracy in altitude is usually the
𝑁( = 5𝐾("! 0 𝐾("# 6 (10) least critical of the three. Rotation will also induce errors at
𝐾(#! 𝐾(#" 0 a rate that is proportional to the rotation rate [5]. Gyro bias
errors accumulate over time as do errors due to alignment
'
0 𝐾)!" 𝐾)!# that spill over into the y and z axes if rotating about x. If 𝜔$'
𝐾 0 𝐾)"# 6 is the angular rate in the sensor frame with respect to the
𝑁) = 5 )"! (11) *
inertial frame, and resolved in the sensor frame, 𝛿𝜔-. is the
𝐾)#! 𝐾)#" 0
error in angular rate due to scale factor in the navigation
*
frame, 𝛿𝜔/ is error in angular rate due to alignment errors in
Rotating about the x-axis, and transferring ns from the sensor the navigation frame, and 𝜔$0 is earth rate with respect to the
frame to the body frame using equations (1) and (6):
inertial frame. The output of the gyros is:
'
1 0 0 1 0 0 𝑛! 𝜔! 𝜔$0 sin 𝐿 + 𝜔
𝑛 ' = 𝐶%* 𝐶'% 𝑛 ' = '0 1 00 5'0 cos 𝜔𝜏 sin 𝜔𝜏 06 5 "' 6
𝑛 '𝜔" 0 = '𝜔$0 cos 𝐿 cos 𝜔𝜏0 (16)
0 0 1 0 − sin 𝜔𝜏 cos 𝜔𝜏 𝑛#' 𝜔# 𝜔$0 cos 𝐿 sin 𝜔𝜏

1 0 0 1 0 0 𝑏!' Where L denotes the local latitude. Error due to scale factor
'
𝑏 = 𝐶%* 𝐶'% 𝑛 ' = '0 1 00 5'0 cos 𝜔𝜏 sin 𝜔𝜏 06 5𝑏"' 6 in the navigation frame can be calculated as follows:
0 0 1 0 − sin 𝜔𝜏 cos 𝜔𝜏 𝑏#'
* '
𝛿𝜔-. = 𝐶'* 𝑁( 𝜔$' = (17)
For this particular case, the navigation frame is not moving;
the IMU will be rotating on a stationary platform about a 𝐾(! 𝜔
single axis for step one, so 𝐶%* is the identity matrix. 5 𝐾(" 𝜔$0 cos 𝐿 cos 𝜔𝜏 cos 𝜔𝜏 + 𝐾(# 𝜔$0 cos 𝐿 sin 𝜔𝜏 sin 𝜔𝜏 6
−𝐾(" 𝜔$0 cos 𝐿 cos 𝜔𝜏 sin 𝜔𝜏 + 𝐾(# 𝜔$0 cos 𝐿 sin 𝜔𝜏 cos 𝜔𝜏
𝑛!'
𝑛 =5 * 𝑛" cos 𝜔𝜏 + 𝑛#' sin 𝜔𝜏 6
'
(12) Using the double angle trigonometry identity and
−𝑛"' sin 𝜔𝜏 + 𝑛#' cos 𝜔𝜏 rearranging terms to:

𝑏!' cos 𝜔𝜏 + 1
(cos 𝜔𝜏)1 =
𝑏 =5 * 𝑏"'
cos 𝜔𝜏 + 𝑛#' sin 𝜔𝜏 6 (13) 2
−𝑏" sin 𝜔𝜏 + 𝑛#' cos 𝜔𝜏
'
sin 2𝜔𝜏
= sin 𝜔𝜏 cos 𝜔𝜏
Integrating nn and bn over one full cycle (2p radians), the y 2
and z terms cancel (since the area above and below the curve − cos 2𝜔𝜏 + 1
are equal), and what’s left is only the x term. (sin 𝜔𝜏)1 =
2
+
; 𝑛* 𝑑𝑡
,

+
; 𝑏* 𝑑𝑡
,

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 5

Substituting: +
I𝐾(!# − 𝐾(#! J𝜔$0 sin 𝐿
𝐾(! 𝜔 *
∫ 𝛿𝜔-. 𝑑𝑡 =5
1
6 (26)
⎡ ⎤ 0
cos 2𝑤𝑡 + 1 − cos 2𝜔𝜏 + 1
⎢𝐾(" 𝜔$0 cos 𝐿 + 𝐾(# 𝜔$0 cos 𝐿 ⎥ 0
⎢ 2 2 ⎥
⎢ sin 2𝜔𝜏 sin 2𝜔𝜏 ⎥ Rotating about the z-axis using the same method provides
⎣ −𝐾(" 𝜔$0 cos 𝐿 + 𝐾(# 𝜔$0 cos 𝐿 ⎦
2 2 similar results:

Integrating: 𝑛!' cos 𝜔𝜏 + 𝑛"' sin 𝜔𝜏


𝑛 = 5 𝑛!' sin 𝜔𝜏 − 𝑛"' cos 𝜔𝜏6
*
(27)
*
; 𝛿𝜔-. 𝑑𝑡 = 𝑛#'
𝐾(! 𝜔
⎡2 ⎤ 𝑏!' cos 𝜔𝜏 + 𝑏"' sin 𝜔𝜏
⎢1 𝑇I𝐾(" + 𝐾(# J𝜔$0 cos 𝐿⎥ (18) 𝑏* = 5 𝑏!' sin 𝜔𝜏 − 𝑏"' cos 𝜔𝜏 6 (28)
⎢2 ⎥ 𝑏#'
⎣ 1 𝑇I𝐾(" + 𝐾(# J𝜔$0 sin 𝐿 ⎦
Integrating:
Rotation rate error due to misalignment can be calculated
similarly (and integrating):
0
0
+ 𝑛* = ' 0 0 (29)
* I𝐾 + 𝐾(#" J𝜔$0 sin 𝐿 M 𝑇𝑛#'
∫ 𝛿𝜔-. 𝑑𝑡 = L 1 ("# (19)
+
I−𝐾("# + 𝐾(#" J𝜔$0 cos 𝐿
1 0
𝑏* = ' 0 0 (30)
Rotating about the y-axis using the same method provides
𝑇𝑛#'
similar results:
For Scale Factor rotating about z:
𝑛!' cos 𝜔𝜏 − 𝑛#' sin 𝜔𝜏
*
𝑛 =5 𝑛"' 6 (20)
I𝐾(! − 𝐾(" J𝜔$0 cos 𝐿 sin 𝜔𝜏 cos 𝜔𝜏
𝑛! sin 𝜔𝜏 + 𝑛#' cos 𝜔𝜏
'
*
𝛿𝜔-. = LI𝐾(! (sin 𝜔𝜏)1 − 𝐾(" (cos 𝜔𝜏)1 J𝜔$0 cos 𝐿M (31)
𝑏!'cos 𝜔𝜏 − sin 𝜔𝜏 𝑏#' 𝐾(# (𝜔$0 sin 𝐿 + 𝜔)
𝑏* = 5 𝑏"' 6 (21)
' '
𝑏! sin 𝜔𝜏 + 𝑏# cos 𝜔𝜏 Integrating:

Integrating: 0
+
*
∫ 𝛿𝜔-. 𝑑𝑡 = 51 I𝐾(! + 𝐾(" J𝜔$0 cos 𝐿6 (32)
0 𝑇𝐾(# (𝜔$0 sin 𝐿 + 𝜔)
𝑛* = '𝑇𝑛"' 0 (22)
0 For alignment errors rotating about z (and integrating):
0 +
𝑏* = '𝑇𝑏"' 0 (23) I𝐾(!" − 𝐾("! J𝜔$0 cos 𝐿
* 1
∫ 𝛿𝜔-. 𝑑𝑡 =5 0 6 (33)
0
0
For Scale Factor rotating about y:
The above equations demonstrate a reduction in error due to
345 167 345 167 noise and bias. Errors still propagate in the axis of rotation.
⎡ −𝐾(! 𝜔$0 sin 𝐿 1 + 𝐾(# 𝜔$0 sin 𝐿 1 ⎤ The above equations also show that IMU rotation induces
*
𝛿𝜔-. =⎢ 𝐾(" (𝜔$0 cos 𝐿 + 𝜔) ⎥ (24) error due to scale factor and alignment. Previous experiments
⎢ 289:3 67 2;9:3 67⎥ [6], [8], [9] demonstrate that the positive aspects of rotation
⎣𝐾(! 𝜔$0 sin 𝐿 1
+ 𝐾(# 𝜔$0 sin 𝐿
1 ⎦
outweigh the negatives.
Single Axis Rotation: Rotation about the z-axis is used in
Integrating:
this technique. Similar techniques were used in other MEMS
0
(𝜔 IMU based rotation experiments [6] [7]. The reference frame
*
∫ 𝛿𝜔-. 𝑑𝑡 = 5 (" $0 cos 𝐿 + 𝜔) 6
𝑇𝐾 (25) is Inertial Tangent [7] using an ENU coordinate system with
+
1
I𝐾(! + 𝐾(# J𝜔$0 sin 𝐿 the y-axis pointing north, the x-axis pointed east, and the z-
axis pointed up. Rotated and un-rotated results will be
For alignment errors rotating about y (and integrating): checked against equations (30), (31), (33) and (35) to
correlate results. The reference [23] experiment conducted in

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

the lab was stationary, and rotated about the z-axis. system used here employs a known 100 rpm 12-volt electric
Navigation time was 5 minutes, and errors built up motors powered with a Nickel-Cadmium battery pack.
substantially over that time period; however, their results Although the rate of each spin motor was measured with a
showed improvement. Reference [8] navigated over a period strobe light for minor variations, spin rates change slightly
of approximately 15 minutes rotating about the x-axis on the while the system is being used. The spin motors warm, and
wheel of a slow-moving land vehicle; however, the z-axis battery pack voltage levels change. Getting stable accurate
was perpendicular to the wheel of the vehicle, and not angular rates from gyroscopes under spin using motors
aligned in the traditional ENU or NED coordinate systems. proved to be more difficult than using an IMU attached to the
Coherent Filters: Coherent filters are used to estimate and wheel of a vehicle [6], [8] where the angular rate is both
compensate for accelerometer and gyro parameter errors that controllable and measurable.
cause significant angular velocity and attitude errors [15]. Spinning the IMU, however, did dramatically reduce the
The accelerometer and gyro parameters that can be modeled bias in angular rates coming off the IMU’s gyroscopes [6].
in the coherent filters are accelerometer and gyro scale Converting from angular rates to angles proved to be
factors, accelerometer and gyro bias, input axis accurate. By the same token, after five minutes, an error of 1
misalignments with respect to the coordinate frame defined to 1.5 degrees was observable making the data unusable.
by the gyro input axes. The concept is to reduce low Consequently, IMU gyroscope data was only used to verify
frequency errors that cannot be calibrated, these are changes in direction, or vehicle turns, and not to establish
deterministic as well as stochastic, and redistribute them accurate body attitude or heading. Gyroscope data helped to
(through rotation) to known frequency bands where divide the data up into sets of data of approximate constant
contribution to position error is small, and can be further heading making acceleration data more useful.
reduced through coherent filtering. See Fig. 1. Accelerometer data Optimization: Acceleration data from
three independent IMUs is available using the concept
described in this paper. Through extensive experimentation
and detailed analysis, it was found that the data set from each
IMU has something that is uniquely useable. Data from one
IMU may have something useful that the other two IMUs
may not, or the useful data is higher quality from one IMU
than another. It’s not as simple as picking the IMU with the
cleanest data, and throwing away the other two. It’s best to
combine the data from all three IMUs in proportional manner
based on the bias or noise observed. For this paper, the data
for walking was similar on all three IMUs. Optimization
offered little improvement. For low-speed vehicle travel,
optimization improved velocity/distance data, but did
nothing for direction. For higher speed vehicle travel,
optimization offered considerable improvement in speed and
direction, and based on bias and noise, a proportion of 20%
for IMU 1, 20% for IMU 2, and 60% for IMU 3 was used.
Calibration: For acceleration bias, the concept for single
axis rotation is to use a simple moving average to estimate
Fig. 1. Error Filtration (Error can be pushed to higher the bias over time in each of three axes, and establish a zero-
frequencies and then filtered). acceleration baseline. The procedure is to turn on the IMU,
and wait for the biases to stabilize while the platform is
Kalman Filters: The system uses a simple Kalman filter
stationary, and enter the results of a simple moving average
for magnetometer data. Where the classic Kalman H and R
with a variable datum based on trip length. The result is
matrices are based on the states being measured. R is based
accurate, and accelerations are useable for 20 minutes or
on measurement error, and H is based on position. The error
more in the navigation solution. However, there is a problem
covariance predicted depends on Q, system noise, that may
with temperature and shock. Shocks to the system may
change based on experimentation, and influences the Kalman
require re-evaluating acceleration biases. Temperature
gain. R, based on measurement noise in the sensor, is
swings will change the biases too. The system can’t be set up
generally left alone. Simple Kalman filters are also used for
in the warm building and moved to a cold or hot vehicle. The
data fusion techniques when a single source is not
biases change, and recalibration is required. The best
dependable over the longer (minutes) time frame.
procedure is to set up and account for the bias in the vehicle.
Allen variance: Describing the connection between Allen
Integrated IMU sensor: IMUs already integrated with
variance and error propagation [8] appears to be very useful
microcontrollers and Serial Data (SD) Card storage devices
in making IMU gyroscope angular rates useable. This could
are employed. Rotations are applied in a single direction
be particularly useful in eliminating one type of noise while
about the gravity vector. Three spinning IMUs are used. An
managing another. Spin rates, however, need to be well
additional non-spinning IMU is used to collect attitude and
controlled in a closed loop. Feedback control adds
magnetometer data. Spinning IMUs are oriented so that each
complexity and cost due to more sophisticated hardware. The

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 7

rotates about a different axis internal to the IMU (x, y, or z). they’re manual or powered, do not come with an odometer
Four IMUs are used. SparkFun DataLogger IoT-9DOF, as for navigation inside large buildings. Wireless bicycle
shown in Fig. 2 are used. The onboard IMU is the odometers can be modified to fit a manual wheelchair.
ISM330DHCX by STMicroelectronics. All data logs to an Compass: Independent compass data and unspun
SD Card in a text file that is imported to MATLAB for magnetometer data is used for direction correlation outdoors.
processing. Besides the integrated SD Card slot, the Most vehicles provide an on-board magnetic based compass.
SparkFun Data Logger IoT-9DoF has an ESP32-WROOM. The compass is only available through a dashboard display,
The WROOM default is Wi-Fi, but requires a data logging and often times requires calibration, and may not be set to
service. The BLE portion of the WROOM is not the proper zone for correction to true North. Handheld
implemented yet. Also notice the LiPo Battery JST-PH 2.0 compasses that record data real time must be calibrated prior
connector. A 3.7V 1000mAh Li-Polymer Battery makes to departure.
each board an independent unit. The IMUs are identical, Pedometers: Pedometers are used for distance estimates for
pedestrian navigation. Calibrating to the user’s stride should
be accomplished each time the pedometer is used.
Setup: Three identical IMUs are rotated by three
independent 12 Volt, 100 rpm (600 degrees per second (dps))
Commercial-Off-The-Shelf (COTS) spin motors, turntable,
and power supply. The power supply is a 12V 2000 milli-
amp hour (mAh) nickel cadmium (ni-cad) battery pack. The
spin motor is shown in Fig. 4.

Fig. 2. IMU Data Logger.

and all powered with their own individual LiPo Battery, and
enclosed in a plastic case to make it easier to attach each one
to spin-platforms. Each has its own power switch, and is
completely independent. See Fig. 3 for IMU, power supply,
switch and case setup.

Fig. 4. Geartisan 12-volt 100 rpm spin motor and ni-cad


battery pack.

The fourth IMU is not rotated, and fixed to the dashboard of


the vehicle with the IMU’s Y axis perpendicular, and the X
axis parallel to the centerline of the vehicle as shown in Fig.
5.

Fig. 3. IMU with LiPo battery, switch and case.

Odometers: Odometers are used in vehicles in order to


correlate distance measurements. Vehicle odometers
generally do not display tenths of a mile; however, the trip Fig. 5 Orientation of the fixed (not rotated) IMU.
odometer will display tenths, and the trip odometer is set
prior to departure and monitored while the vehicle is in For the rotating IMUs, the spin axis of all three motors is
motion. Many manufactures will provide odometer data as parallel to the gravity vector. The spin axis of each IMU is
part of the On-Board Diagnostics (OBD) [24], but neither different. IMU 1 spins about the X-axis, and X-axis
OBD-1 nor OBD-2 will display odometer data while the coincides with the gravity vector. IMU 2 spins about the Y
vehicle is running. Unfortunately, wheelchairs, whether axis, and the Y-axis coincides with the gravity vector.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

Finally, IMU 3 spins about the negative Z axis, and the Z-


axis coincides with the gravity vector. The switches turn on
and off each spin motor. Each motor has its own individual
12-volt ni-cad battery pack. The spin motor/IMU setup is
shown in Fig. 6.

Fig. 7. Test Fixture. For rotating each of the three


IMUs. Mounting tables spin at 100 rpm (600 dps).

Fig. 6. Spin motor/IMU setup.


Test Fixture: The three IMUs and their spin motors are
mounted on a custom-made bracket that attaches to a
computer monitor stand. There is a separate power switch for
each spin motor. The switch assembly is also attached to the
custom bracket. The three 12-volt battery packs are strapped
to the base of the stand, and require occasional charging. The Fig. 8. Shock absorbing material for Test Fixture.
test fixture is shown in Fig. 7. The bottom of the test fixture
is fitted with shock absorbing material for shock isolation. Path 1, Short
250
The shock absorbing material is shown in Fig. 8.
GPS Reference: Global Positioning System (GPS) with
Real Time Kinematics (RTK) is used for position reference 200
if needed. Specifically, the Spark Fun GPS-RTK Dead
Distance North (meters)

Reckoning Breakout-ZED-F9R provides resolution down to


one meter without subscribing to support services. 150

Test Path: Inertial navigation is performed on the separate


test paths shown in Fig. 9 and Fig. 10. Path 1, Fig. 9, takes 100
approximately 50 seconds at vehicle speeds of 24.1 kph (15
mph). At walking speeds (approximately 4.8 kph (3 mph)),
the path is an approximate 360 second (6-minute) walk. Path 50

2, Fig. 10, is traversed at freeway speeds of approximately


104.6 kmph (65 mph). The measured path will be plotted 0
with the actual path. -180 -160 -140 -120 -100 -80 -60 -40 -20 0
Distance East (meters)

Fig. 9. Short local 286.3-meter (.18 mile) path.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 9

Path 2, Long Optimized: Acceleration X and Y


1
5000 X
0.8 Y

0.6
0
Distance North (meters)

0.4

Acceleration (m/sec 2 )
0.2
-5000
0

-0.2
-10000
-0.4

-0.6
-15000
-0.8

-1
-2000 0 2000 4000 6000 8000 10000 12000 0 100 200 300 400 500 600 700 800
Distance East (meters) Time (tenths of seconds)

Fig. 10. Long path, 30,480-meter (19.5 mile). Fig. 11. Optimized acceleration data proportionally blending
data from the three rotating IMUs.
III. Results
Vehicle Short Path: Using the proportion 20%, 20%, 60% 1.5
IMU 2: Acceleration X and Y

previously described for optimizing the acceleration outputs X


Y
for the three spinning IMUs, the optimized outputs in Fig. 11 1
shows a significant improvement from the output of a single
IMU shown in Fig. 12. At vehicle speeds of approximately
Acceleration (m/sec 2 )
0.5
24.1 kph (15 mph), accelerometer data becomes useable for
distance, but not for direction. It can be seen in Fig. 12 that
0
error increases with time. With only 50 seconds of drive
time; however, accelerometer error is in an acceptable range
for distance. The direction, however, was taken from -0.5

magnetometer readings. The direction was correlated to the


vehicle’s compass data, but no corrections were necessary. -1

Although not shown, direction based on acceleration data


was unusable. Distance traversed, based on accelerometer -1.5
0 100 200 300 400 500 600 700 800
data, was accurate [25]. The plot of Fig 13 shows a combined Time (tenths of seconds)
solution of distance from accelerometer data and direction Fig. 12. Un-optimized acceleration data from IMU 2 only.
from magnetometer data. Magnetometer data can be subject
to error due to large steel structures such as bridges or high 57 Seconds of Driving Time
voltage power lines, but in most outdoor applications, 250
Rest to 6.25 m/sec to rest

magnetometer and compass data work fine for direction. Fig. Actual
Measured
14 shows the acceleration plots before the fixed bias is
calibrated out. Notice how bias corrects after spin is turned 200

on at approximately 1000 tenths of seconds. This is the full


Distance North (meters)

set of data going and returning. The first pass is from 2011 150
to 2586 tenths of seconds. The return path is from 3188 to
3855 tenths of seconds. The data for each run is very similar
but the bias is different. Only the going data was used. Fig. 100

15 shows the magnetometer data. Notice how the


magnetometer data swaps symmetrically for the going and 50
return paths.
0
-180 -160 -140 -120 -100 -80 -60 -40 -20 0
Distance East (meters)

Fig. 13. Results of Vehicle short path. Error increases with


time. Direction based on magnetometer data and distance
based on accelerometer data.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

Acceleration to earth rate and gravity fluctuations [28]. Larger


10
X acceleration readings can be used to denote the beginnings
8 Y and endings of changes in direction [29]. It’s easier to capture
6
the large acceleration readings and relate them to the
magnetometer data for a heading vector. For overall
acceleration data, IMU noise remains a factor. Heading
Acceleration (m/sec 2 )

2
vectors obtained from pure acceleration are frequently
inaccurate and often inconsistent between runs.
0
Magnetometer readings correlated to acceleration changes
-2 are consistent. Large accelerations can be used to parse the
magnetometer data into direction bins, filter the
-4
magnetometer data within those bins, and then use that as the
-6 sole heading vector.
-8
0 1000 2000 3000 4000 5000 Filtered Magnetometer-direction, Acceleration-distance
Time (tenths of seconds) 250
Actual
Fig. 14. Results of Vehicle short path, accelerations 2011 to Measured

2586 tenths of seconds going, and 3188 to 3855 tenths of 200


seconds returning.

North Distance (meters)


Magnetometer
0.5
X 150
Y
0.4

100
Earth's Magnetic Field (gauss)

0.3

0.2
50
0.1

0 0
-150 -100 -50 0
East Distance (meters)
-0.1
Fig. 16. Short Walking Path Results, Navigating the short
-0.2 path walking using magnetometer data filtered.
-0.3
0 1000 2000 3000 4000 5000 Unfiltered Magnetometer-direction, Acceleration-distance
Time (tenths of seconds) 250
Actual
Fig. 15. Results of Vehicle short path, magnetometer Measured

readings, 1868 to 2416 tenths of seconds going, and 3148 to 200


3683 returning.
Distance North (meters)

Walking Short Path: With the same path slightly modified 150
for the sidewalk, Fig 16 and Fig 17 show the results at
walking speeds. At walking speeds, accelerometer data from
100
the IMUs is less accurate. It’s not just low speeds that
degrades accuracy. The instability of the test fixture while
walking adds error. Data from the rotating IMUs becomes 50
unusable, and the fixed IMU becomes the primary data
source if mounted on the belt as in Ref. [26]. Accelerations
drift off at walking speeds after approximately 3.5 minutes. 0
-180 -160 -140 -120 -100 -80 -60 -40 -20 0
After the first turn, distance data is correlated to the Distance East (meters)

pedometer for accuracy. Further improvement on direction Fig. 17 Short Walking Path Results, Navigating the short
was obtained after passing the data through a simple Kalman path walking using magnetometer data unfiltered.
Filter. Fig. 18 shows accelerometer data and Fig. 19
magnetometer data. System noise from the IMU has a
tendency to cover small acceleration readings. When
walking, speed is fairly constant, and there are short bursts
of acceleration when starting and stopping related to speed.
These short bursts of acceleration are hard to capture within
the IMU noise, and any type of filtering (moving average,
Kalman [15], or otherwise [27]), doesn’t always help.
Having a fixed local reference eliminates complications due

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 11

Acceleration Optimized: Acceleration X and Y


15 1.5
Xacc X
Yacc Y
10
1

5
Acceleration (m/sec 2 )

Acceleration (m/sec 2 )
0.5
0

-5 0

-10
-0.5

-15

-1
-20

-25 -1.5
0 1000 2000 3000 4000 5000 6000 0 500 1000 1500 2000 2500 3000
Time (tenths of seconds) Time (tenths of seconds)

Fig. 18. Acceleration data drifted off after the first turn at Fig. 20. Optimized acceleration data proportionally blending
approximately 3100 tenths of seconds. data from the three rotating IMUs.

Magnetometer IMU 2: Acceleration X and Y


0.4 2
Xmag X
Ymag Y
1.5
0.2

1
Earth's Magnetic Field (gaus)

Acceleration (m/sec 2 )
0
0.5

-0.2 0

-0.5
-0.4

-1

-0.6
-1.5

-0.8 -2
0 1000 2000 3000 4000 5000 6000 0 500 1000 1500 2000 2500 3000
Time (tenths of seconds) Time (tenths of seconds)

Fig. 19. Magnetometer data shows a spike in Y at the same Fig. 21. Un-optimized acceleration data from IMU 2 only.
time as acceleration data but recovers.
Magnetometer data is the most accurate for direction but
Vehicle Long Path: Optimizing the acceleration data from shifting to acceleration data for distance degrades the
three separate IMUs was very successful at the higher speeds accuracy of the 3rd plot slightly. If acceleration data is used
and longer transit times of the long path. Speeds of for both distance and heading as shown in plot 4, there is
approximately 104.6 kilometers/hour (kph) (65 mph) were further degradation, but remains useable, and shows
maintained for the trip. The trip is 20 miles long and takes excellent results for a 20-minute trip time. It is evident from
approximately 20 minutes including the lower speeds prior the bowed shape of the second freeway leg of the fourth plot
to entering and after exiting the freeway. The beginning and that direction drifts off to the point of being one mile off at
ending of the drive are at surface street speeds of its maximum. It is also evident from the beginning of the 4th
approximately 64.4 kph (40 mph). Fig. 20 shows optimized plot, that acceleration data is not handling direction well
acceleration data using all three IMUs. Un-optimized data when not on the freeway. Through experimentation, drifting
from IMU 2 is shown in Fig. 21. In order to show the detail, direction data seems to be related to vehicle turns. There are
only the first portion of the data is shown, but all plots show several turns in the beginning of the plot. At the higher 100
similar results. Figure 22 shows the results of the navigation rpm spin rate, and at higher vehicle speeds, direction based
solution using optimized acceleration data blended from all on acceleration data recovers with time, and an accurate
three rotating IMUs. Four plots are shown. The first is the endpoint can be established. Better correlation between data
actual path. The second is using odometer data for distance sources may further improve plot 4. Fig. 23 shows raw
and magnetometer data for direction. The third is using acceleration data that has a high bias of 0.25 m/sec2 during
accelerometer data for distance and magnetometer data for periods of constant velocity on the freeway portions of the
direction, and the fourth is using accelerometer data from all drive. Fortunately, this can be calibrated out. What’s more
three spinning IMUs for both distance and direction. It is concerning is the bias change in Y at approximately 5000
evident that the most accurate plot uses odometer data from tenths of seconds. Fig. 24 shows magnetometer data that
the vehicle combined with magnetometer data for direction. tracks the actual direction well without filters. Table 1

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

summarizes the improvement as a result of using three Magnetometer Readings


1.5
independent rotating IMUs (and one fixed IMU) to optimize X
the quality of the output data. Y
1

Table 1: Blending output of 3-IMUs to reduce bias

Earth's Magnetic Field (gauss)


Mean Mean Mean 0.5

Bias Distance Direction


Reduction Accuracy Accuracy 0

Improvement Improvement
24 kph 38% 12% 0% -0.5
(15 mph)
104.6 -1
kph 42% 14% 73%
(65 mph) -1.5
0 5000 10000 15000
Time (tenths of seconds)
20 miles, 20 min, mostly steady 104.6 kmh (65 mph)
2000 Fig. 24. Vehicle Long Path magnetron data. Direction
0
tracks well with the actual path.
-2000
III. CONCLUSIONS
Distance North (meters)

-4000
Proportionally combining data from the output of three
-6000 independent and identical IMUs can significantly improve
-8000
heading and velocity results. An improved dead reckoning
navigation solution results during periods of GPS denial.
-10000
Multiple sources have demonstrated the advantage of
-12000 rotating an IMU in the past [6], [8] and [9]. Using combined
-14000
and optimized results of multiple rotating IMUs takes 20-
minute dead reckoning to a new level.
-16000 Optimized IMU data can be used on its own, but can be
0 2000 4000 6000 8000 10000 12000 14000 further improved through correlation to odometry and
Distance East (meters)
magnetometry. Supplementary systems are low cost, useful,
Fig. 22. Vehicle Long Path Results. Accelerometer data and should not be overlooked. The addition of multiple
becomes useful for both direction and distance (purple). rotating IMUs combined with navigation aids would make
solid navigation possible in all environments. Standalone
2.5
Accelerations Measured speedometers, odometers, and magnetometers are readily
X available. In addition, navigation aids combined with
Y
2 mapping and SLAM techniques could provide a navigation
1.5 solution for wheelchairs and other personal mobility devices
indoors.
Acceleration (meters/sec 2 )

1
Although the concepts provided here offer a solution,
0.5 some problems remain. Shock and temperature swings are a
challenge for bias stability. Frequent calibration is required.
0
The calibration process is complex, and needs improvement
-0.5 through simplification. Odometer readings in some vehicles
-1
can be difficult to get. The trip odometer is often the only
measure. Resolution of a trip odometer is only one tenth of a
-1.5 mile. Vehicle manufacturers should add the capability of
-2 broadcasting data, through BLE, to include odometry and
0 2000 4000 6000 8000 10000 12000
Time (tenths of seconds)
14000 16000
magnetometry data.
With minimum changes to the way modern navigation is
Fig. 23. Vehicle Long Path raw accelerations. There’s
treated, the consumer can benefit by supplementing GPS
significant bias during periods of constant velocity.
based systems. Proof of concept, has been established
through the multiple independent rotating IMU techniques
presented here, as well as other past experiments as described
in reference [30]. Navigating, primarily with IMUs, as well
as all available local sensors, fills the navigation gap, and
completes the solution.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/
This article has been accepted for publication in IEEE Open Journal of Instrumentation and Measurement. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/OJIM.2024.3477574

OJIM-24-00056 13

References [22] Kim, Phil, Rigid Body Dynamics, Euler Angles and Quaternions,
Published by Phil Kim, 2013, page 32.
[23] Du, Shuang, Sun, Wei, and Gao, Yang, MEMS IMU Error Mitigation
[1] Jenks, Shepard H., Navigating Under the North Pole Icecap, U.S. Using Rotation Modulation Techniques, Sensors 2016, 16(12), 2017,
Naval Institute Proceedings, Vol. 84/12/670, December 1958. August 6, 2016.
[2] Time and Navigation, The Untold Story of Getting from Here to There, [24] “Your car is watching you. Who owns the data?”, Roll Call, April 9,
Shipboard Inertial Navigation System (SINS), The Smithsonian 2019.
Institution, National Air and Space Museum NASM2013-00298, [25] Rogne, Robert H., Bryne, Torleiv H., Fossen, Thor I., Johansen, Tor
December 20, 2012. A., MEMS-based Inertial Navigation on Dynamically Positioned
[3] Compare our Inertial Measurement Units, Honeywell, Ships: Dead Reckoning, Science Direct, IFAC Papers, 49-23 (2016)
https://aerospace.honeywell.com/us/en/products-and- 139-146.
services/product/hardware-and-systems/sensors/inertial- [26] Hajati, Nasim, Rezaeizadeh, Amin, A Wearable Pedestrian
measurement-units, 2024. Localization and Gait Identification System Using Kalman Filtered
[4] Geller, Edward S., Inertial Platform System Rotation, Sperry Inertial Data, IEEE Transactions on Instrumentation and
Gyroscope Company, IEEE Transactions on Aerospace and Electronic Measurement, Vol. 70 , 2021.
Systems, VOL. AES-4, No. 4, July 1968. [27] Madgwick, Sebastian O.H., An Efficient Orientation Filter for Inertial
[5] Walsh, Ed, Navy and industry investigate new super-accurate optical and Inertial/Magnetic Sensor Arrays, April 30, 2010.
gyros for possible use on ballistic missile submarines, Military + [28] Britting, Kenneth, R, Inertial Navigation System Analysis,
Aerospace Electronics, November 30,2001. Massachusetts Institute of Technology, Cambridge, MA, Wiley and
[6] Niu, Xiaoji, Wu, Yibin, Kuang, Jian, Wheel-INS: A Wheel-Mounted Sons, 1971, pages 30-43.
MEMS IMU Based Dead Reckoning System, IEEE Transactions on [29] Eskin, Maksim, Design of an Inertial Navigation Unit Using MEMS
Vehicular Technology, Vol.70 No. 10, October 2021. Sensors, A Design Project Report Presented to the Engineering
[7] Collin, Jussi, MEMS IMU Carouseling for Ground Vehicles, IEEE Division of the Graduate School of Cornell University in Partial
Transactions on Vehicular Technology, Vol. 64, No. 6, June 2015. Fulfillment of the Requirements for the Degree of Master of
[8] Collin, Jussi, Kirkko-Jaakkola, Martti, Takala, Jarmo, Effect of Engineering (Electrical), January 2006.
Carouseling on Angular Rate Sensor Error Processes, IEEE [30] Collin, Jussi, Mezentsev, Oleg, Design Performance of Wheel-
Transactions on Instrumentation and Measurement, Vol. 64 No.1, mounted MEMS IMU for Vehicular Navigation, IEEE, 2019, 6th
January 2015. Symposium on Inertial Sensors and Systems.
[9] Wang,Xueyun, Wu, Jie, Xu, Tao, Wang, Wei, Analysis and
Verification of Rotation Modulation Effects on Inertial Navigation
Systems Based on MEMS Sensors, The Journal of Navigation, The
Royal Institute of Navigation, Vol 66, 2013.
[10] Morris, Cameron, Vedang, Chauhan, Design of Low-Cost
Autonomous Robot for Outdoor Applications, Proc. ASME.
IMECE2022, Volume 5, October 30–November 3, 2022.
[11] SP 24 Navigation Familiarization, Strategic Systems Programs, 2015.
[12] Keller, John, Northrop to provide legacy AN/WSN-7 ring laser gyro
inertial navigation for shipboard electronics, Military Aerospace and
Electronics, December 11, 2023.
[13] El-Diasty M., El-Rabbany, A., Pagiatakis, S., Temperature Variation
Effects on Stochastic Characteristics for low-cost MEMS-based
inertial sensor error. Measurement Science and Technology,
Measurement Science and Technology, Volume 18 No. 11, September
20t, 2007.
[14] Sharkov, A., Vinogradov, A.V., Kosolov, V.N., Strigalev, V.E.,
Kikilich, V.N., Atmospheric Pressure Effect on the Fiber Optic
Gyroscope Output Signal, Information Technologies, Mechanics and
Optics, Volume 17 No. 3, May-June 2017.
[15] Kim, Phil, Kalman Filter for Beginners with MATLAB Examples,
Published by Phil Kim, 2011, pages 131 to 140.
[16] Avzayesh, Mohammad, Al-Masri, Wasim, Abdel-Hafez, Mamoun F.,
Alshabi, Mohammad, Improved-Performance Vehicle’s State
Estimator Under Uncertain Model Dynamics, IEEE Instrumentation
and Measurement, March 25, 2024.
[17] Allotta, Benedetto, Costanzi, Riccardo, Meli, Enrico, Ridolfi,
Alessandro, Chisci, Luigi, Fantacci, Claudio, Caiti, Andrea, Di
Carato, Frencesco, Fenucci, Davide, An Innovative Navigation
Strategy for Autonomous Underwater Vehicles: An Unscented
Kalman Filter Based Approach, ASME Proceedings, IDETC-
CIE2015, Volume 5A, August 2–5, 2015.
[18] Carey, Kevin, Abruzzo, Benjamin, Harvie, David P., Korpela,
Christopher, Performance Comparison of Inertial Measurement Units
Fused with Odometry in Extended Kalman Filter for Dead-
Patrick Grates (M’22) received a B.Sc. in
Reckoning Navigation, ASME Proceedings, IDETC-CIE2019,
Mechanical Engineering from Michigan Tech
Volume 5A, August 18–21, 2019.
University 1981, and a M.Sc. in Systems
[19] Wang, Tianfu, Asokanthan, Samuel F., Dynamic Stability of a Single-
Engineering from Naval Post Graduate School,
Axis Vibratory MEMS Gyroscope, ASME Proceedings, IDETC-
Monterey, CA.
CIE2005, Volume 1, September 24–28, 2005.
Mr. Grates is currently a PhD. Student in
[20] Redkar, Sangram, Sugar, Tom, Dillard, Bill, Narayanan, Karthik,
Systems Engineering at Arizona State University.
Inertial Sensing of Dummy Kinematics, ASME Proceedings. IDETC-
His research is in low-cost navigation.
CIE2009, Volume 6, August 30–September 2, 2009.
Member Institute of Navigation (ION) and
[21] Rogne, Robert H., Byrne, Torleiv, H., Fossen, Thor. I., Johansen, Tor.
ASME. Registered Mechanical Engineer, State of
H., Redundant MEMS-Based Inertial Navigation Using Nonlinear
California, License 28901.
Observers, ASME Journal of Dynamic Systems, Measurements and
Control, July, 2018.

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

You might also like