sensors

Article
Robust Attitude Estimation for Low-Dynamic Vehicles Based on
MEMS-IMU and External Acceleration Compensation
Jiaxuan Chen 1,2 , Bingbo Cui 1,2, * , Xinhua Wei 1,2 , Yongyun Zhu 1,2 , Zeyu Sun 1,2 and Yufei Liu 3

1 Key Laboratory of Modern Agricultural Equipment and Technology, Jiangsu University, Ministry of
Education, Zhenjiang 212013, China; [email protected] (J.C.); [email protected] (X.W.);
[email protected] (Y.Z.); [email protected] (Z.S.)
2 School of Agriculture Engineering, Jiangsu University, Zhenjiang 212013, China
3 College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China;
[email protected]
* Correspondence: [email protected]

Abstract: Attitude determination based on a micro-electro-mechanical system inertial measurement

unit (MEMS-IMU) has attracted extensive attention. The non-gravitational components of the MEMS-
IMU have a significant effect on the accuracy of attitude estimation. To improve the attitude estimation
of low-dynamic vehicles under uneven soil conditions or vibrations, a robust Kalman filter (RKF)
was developed and tested in this paper, where the noise covariance was adaptively changed to
compensate for the external acceleration of the vehicle. The state model for MEMS-IMU attitude
estimation was initially constructed using a simplified direction cosine matrix. Subsequently, the
variance of unmodeled external acceleration was estimated online based on filtering innovations
of different window lengths, where the acceleration disturbance was addressed by tradeoffs in
time-delay and prescribed computation cost. The effectiveness of the RKF was validated through
experiments using a three-axis turntable, an automatic vehicle, and a tractor tillage test. The turntable
experiment demonstrated that the angle result of the RKF was 0.051◦ in terms of root mean square
error (RMSE), showing improvements of 65.5% and 29.2% over a conventional KF and MTi-300,
respectively. The dynamic attitude estimation of the automatic vehicle showed that the RKF achieves
Citation: Chen, J.; Cui, B.; Wei, X.;
smoother pitch angles than the KF when the vehicle passes over speed bumps at different speeds; the
Zhu, Y.; Sun, Z.; Liu, Y. Robust RMSE of pitch was reduced from 0.875◦ to 0.460◦ and presented a similar attitude trend to the MTi-
Attitude Estimation for Low-Dynamic 300. The tractor tillage test indicated that the RMSE of plough pitch was improved from 0.493◦ with
Vehicles Based on MEMS-IMU and the KF to 0.259◦ with the RKF, an enhancement of approximately 47.5%, illustrating the superiority
External Acceleration Compensation. of the RKF in suppressing the external acceleration disturbances of IMU-based attitude estimation.
Sensors 2024, 24, 4623. https://
doi.org/10.3390/s24144623 Keywords: dynamic attitude estimation; inertial measurement unit; robust Kalman filter; external
Academic Editor: Arturo de la
acceleration compensation
Escalera Hueso

Received: 23 June 2024

Revised: 13 July 2024
1. Introduction
Accepted: 15 July 2024
Published: 17 July 2024
Accurate attitude information not only plays a crucial role in vehicle navigation and
localization, but is also an important parameter for operational control of agricultural
implements, such as monitoring and controlling tractor tillage depth [1–3]. Generally, the
variation of vehicle attitude can be updated by integrating the output of gyroscope angular
Copyright: © 2024 by the authors. rates. However, lack of knowledge of the initial vehicle attitude and gyroscope bias may
Licensee MDPI, Basel, Switzerland. lead to drift in attitude calculation. This issue is particularly significant when dealing with
This article is an open access article micro-electro-mechanical system (MEMS) gyroscopes, as their noise characteristics and
distributed under the terms and environmental drift errors are highly complicated. Based on the accelerometer’s specific
conditions of the Creative Commons force measurements, calculation of attitude angles can be achieved under static and uniform
Attribution (CC BY) license (https:// motion conditions of the vehicle. However, the measurements are affected by external
creativecommons.org/licenses/by/ accelerations [4–6]. Attitude estimation based on a low-cost inertial measurement unit
4.0/).

Sensors 2024, 24, 4623. https://doi.org/10.3390/s24144623 https://www.mdpi.com/journal/sensors

Sensors 2024, 24, 4623 2 of 16

(IMU) has attracted significant attention due to the complementary error characteristics of
gyroscope and accelerometer in calculating the vehicle attitude [7–9].
Popular attitude estimation methods include complementary filter (CF) [10,11] and
extended Kalman filter (EKF) [12,13]. The CF method utilizes frequency complementary
features of the gyroscope-derived and accelerometer-derived attitude. The EKF method
constructs a measurement model based on the difference between the accelerometer’s
observations and the projection of gravitational effects on the vehicle body axis, and it
updates the attitude in the time domain based on the gyroscope output. While the CF
method is characterized by reduced computational complexity compared to the Kalman
filter (KF), its rate of convergence is impacted by the initial attitude estimation [14]. Fur-
thermore, the KF exhibits greater flexibility compared to the CF in model design, especially
in the case of dynamic conditions. The KF method and its variants have found widespread
applications in the field of autonomous driving [15–19]. With the increase of computa-
tional capabilities for microcontrollers, there is a growing advantage in exploring KF-based
methods for attitude estimation in navigation and control applications [20–22]. Many
robust KFs have been proposed for the measurement of outlier detection in GNSS-based
navigation [23,24], which however cannot be applied in IMU-based attitude estimation
due to the attitude modulation of accelerometer observations. Javed et al. proposed a
cascaded KFs structure to compensate gyroscope biases and imported an adaptive external
acceleration model to detect external acceleration [25]. Odry et al. utilized IF-THEN rule-
based adaptation laws to adjust noise covariance matrices dynamically, which suppressed
external disturbances effectively [26]. The above-mentioned KF-based attitude estimators
primarily rely on adaptive measurement noise setting to modify the Kalman gains once
external acceleration is detected. However, there is a distinct difference between external
acceleration and accelerometer noise, which is that the modeling error of external accelera-
tion is scenario-dependent and cannot be performed once for all conditions. Consequently,
further investigation is required to improve the accuracy of attitude control under different
dynamic conditions.
In the navigation of agricultural machinery and operational control of implements,
KF-based attitude estimation has been widely employed to correct positioning errors or
improve implement control stability [27,28]. Huang et al. developed a precise tilt angle
monitor system by using the Euler angle algorithm to calculate attitude, where the noise
covariances were adjusted adaptively to handle time-varying working conditions [29].
Yu et al. improved the implement leveling control system by reducing the implement
attitude estimation error based on combined information from gyroscope and accelerom-
eter [30]. Yang et al. proposed a pitch angle prediction model for tractors based on time
series analysis and an EKF, which improved the dynamic response speed of tilling depth
regulation [31]. Zhao et al. proposed an improved adaptive Kalman filtering method for
sowing depth detection after analyzing the movement mechanism of a seed drill’s parallel
four-bar linkage [32]. All of these attitude prediction model-based control systems share a
common premise: that the attitude estimation model can address the actual attitude of the
agricultural implement, and external accelerations are neglected. However, in the case of
frequent vibration and uneven soil conditions for heavy machinery, such as in the design of
tractor tillage depth control systems, this premise is not true and the external acceleration
may negatively affect the attitude estimation.
Recently, Candan et al. proposed an adaptive external acceleration compensation
method based on KF innovations, where a diagonal weight matrix was designed for triaxial
accelerometers, and the method was verified using IMU data from an unmanned aerial
vehicle (UAV) [33]. In this paper, we further validate the robust KF (RKF) for attitude
estimation of low-dynamic vehicles, such as an automatic control vehicle and tractor, where
the window length selection for attitude estimation under different dynamic conditions is
investigated. Experiments based on turntable tests, automatic control vehicles and tractor
tillage field tests were employed to evaluate the performance of the RKF. The experimental
Sensors 2024, 24, 4623 3 of 16

results can provide a valuable reference on how to compensate for external acceleration in
attitude determination of a low-dynamic vehicle.
The structure of this paper is arranged as follows. The simplified attitude filter model is
briefly reviewed in Section 2, and then the robust KF-based attitude estimation is presented.
In Section 3, the RKF is verified by employing a turntable experiment, automatic vehicle
and tractor field test. Finally, Section 4 concludes this work.

2. Materials and Methods

2.1. KF Algorithm
The KF algorithm consists of time-updating and measurement-updating stages. The
discrete state-space model (SSM) can be written as follows:

xk = Φk−1 xk−1 + wk−1 (1)

zk = Hk xk + vk (2)
where xk is the system state at time k; Φk−1 is the system state transition matrix; zk is the
system measurement vector; Hk is the system measurement matrix; while wk−1 and vk are
the system noise and measurement noise, respectively; the initial state satisfies N (x0 , P0 )
which is independent with wk−1 and vk , where N (x0 , P0 ) denotes a Gaussian distribution
with mean x0 and variance P0 . The KF time update is formulated as [15]

x−
k = Φk −1 xk −1 (3)

P− T
k = Φk −1 Pk −1 Φk−1 + Qk −1 (4)
where x− −
k , Pk are the prediction state and corresponding covariance matrix, and Qk −1 is
the system noise covariance matrix. The measurement update of the KF can be written as
  −1
Kk = P− H
k k
T
H k P − T
H
k k + R k (5)

xk = x−
k + Kk ek (6)
Pk = (I − Kk Hk )P−
k (7)
where Kk is KF gain; ek = zk − Hk x− the filtered innovation; I is the identity matrix
k is 
of proportional dimension; and Rk = E vk vkT is the noise covariance matrix. The KF


operates under the assumption that the state estimation at time k depends solely on the
state at the previous time and the measurement value at the current time. Through iterative
processes, it estimates the state and its uncertainty in the temporal domain, achieving
optimal estimation when Gaussian noise is assumed in the linear state space.

2.2. IMU Attitude Estimation

2.2.1. Principle of Attitude Calculation
The angular rate and specific force of the vehicle frame (b-frame) relative to the
inertial frame (i-frame) are measured by the gyroscopes and accelerometers of the IMU.
Calculating the vehicle’s attitude angle under dynamic conditions involves solving the
attitude differential equations using the outputs of the gyroscopes and transforming them
into the navigation frame (n-frame). In this work, the navigation frame is selected as
North-East-Down, and the vehicle frame is set as Front-Right-Down. Popular attitude
updating methods include direction cosine matrix (DCM) and quaternion updating. In this
study, a simplified DCM is employed to determine the vehicle’s roll and pitch angles. The
transformation matrix from b-frame to n-frame can be formulated as
 
cαcβ cαsβsγ − sαcγ cαsβcγ + sαsγ
Cbn =  sαcβ sαsβsγ + cαcγ cαsβsγ − cαsγ (8)
−sβ cβsγ cβcγ
Sensors 2024, 24, 4623 4 of 16

where s denotes the sine function; c denotes the cosine function; α, β, and γ are the rotation
angles of the b-frame around the Z, Y, and X axis, respectively. Consequently, the pitch and
roll angles are defined as follows:
 
− C 31
β = tan−1  q  (9)
2 2
C33 + C32
 
−1 C32
γ = tan (10)
C33
where Cij is the element of matrix Cbn at the ith row and jth column. Due to the existence of
gravity, based on Equation (8) the projection of gravity to b-frame gravity can be expressed
as follows:  
−sβ
T
gb = (Cbn ) gn =  cβsγ · ge (11)
cβcγ
where ge is the local gravitational acceleration. In Equations (6)–(8), the last row of Cbn not
only calculates β and γ, but can also be used to compensate for the effect of acceleration
measurements ge , thereby the state vector of attitude SSM is selected as

xk = (C31 , C32 , C33 ) T (12)

2.2.2. Filtering Model Construction

The gyroscope and accelerometer measurements at time k can be expressed as follows:
b
yg,k = ωib,k + ng,k (13)

ya,k = ak + gb + na,k (14)

where yg,k and ya,k are the outputs of the gyroscope and accelerometer; ωib,kb and a are the
k
ideal angular rate and external acceleration; while ng,k and na,k are the zero-mean Gaussian
white noise of the sensor. Suppose the biases of gyroscope and accelerometer have been
compensated and we can thus ignore the random bias error in (13) and (14). The external
acceleration is assumed to change slowly and can be modeled using a first-order low-pass
filtered white noise process.
ak = c a ak −1 + ε k (15)
where c a is a constant ranging from 0 to 1 and εk is the time-varying error in modeling
the vehicle’s acceleration motion. The first-order approximation is employed to solve the
differential equation of DCM at time k, then we have
  
Cnb,k = Cnb,k−1 I3 + ∆T ωib,k
b
× (16)

b is the ideal measurement of the gyroscope; ( ω ×) denotes the skew-symmetric

where ωib,k
matrix of the vector ω, and ∆T is the sampling period of the gyroscope. It is notable that
b is neglected in deriving (16), as the gyroscope may not be sufficiently
the earth rate ωie,k
accurate to distinguish earth rate from random noise. Substituting the actual measurement
from Equation (13) into the above equation and corresponding to Equation (1) yields the
following results:  
Φk−1 = I3 − ∆T yg,k × (17)

wk−1 = ∆T (−xk−1 ×)ng,k (18)

Subsequently, the system noise covariance matrix is expressed as

Qk−1 = −∆T 2 (xk−1 ×)Σ gg (xk−1 ×) (19)

Sensors 2024, 24, 4623 5 of 16

where Σ gg is the noise covariance of the gyroscope measurement. By substituting Equation (15)
into Equation (14) and associating it with Equation (11), we have

ya,k − c a ak−1 = ge xk + εk + na,k (20)

Corresponding to Equation (2), the following expressions can be obtained:

zk = ya,k − c a ak−1 (21)

Hk = ge I3 (22)
vk = εk + na,k (23)
Since εk and na,k are uncorrelated noise terms, we have

Rk = Σ acc,k + Σ aa (24)

where Σ aa is the accelerometer noise covariance and Σ acc,k is the corresponding variance
of the external acceleration modeling error. Based on Equations (17)–(24), a simplified
DCM attitude estimation model can be derived. In this model, the external acceleration
motion modeling error of the mobile vehicle is addressed by Rk , which affects the atti-
tude estimation results and must be suppressed during dynamic attitude estimation of
the vehicle.

2.3. Robust Attitude Estimation Method

Attitude estimation based on adaptive measurement noise has been applied in UAVs [33],
whereas its performance in processing vibration for low-dynamic vehicles is seldom re-
ported. This study focuses on measurement anomaly detection and an online noise estima-
tion approach based on KF innovation. It also aims to compensate for measurement model
uncertainty and assess the impact of its application on attitude estimation in the presence
of vibration interference on low-dynamic vehicles. The innovation of the KF denotes the
deviation between predicted and actual measurements. It accurately reflects the deviation
of the estimated state from the actual underlying state, with its covariance defined as the
covariance between predicted and actual measurements. The innovation covariance of the
KF is expressed as h i
Ck = E ek ekT = Hk P− T
k Hk + Σ̂ acc,k + Σ aa (25)

Subsequently, Kalman gain matrix is determined as

  −1
Kk = P− T − T
k Hk Hk Pk Hk + Σ̂ acc,k + Σ aa (26)

where Σ̂ acc,k is the online estimated value at moment k. When there is an external acceler-
ation for the vehicle, Σ̂ acc increases, Kk decreases, and xk is more dependent on the time
update of the KF. To minimize the computational load of the RKF, the variance of the
noise in the measurement associated with external acceleration is updated only when a
substantial deviation exists between the trace value of detected and predicted innovation
covariance, that is    
tr ek ek T > tr Hk P− H
k k
T
+ Σ̂ acc,k + Σ aa (27)

where tr(·) is the operation of matrix trace calculation. It is assumed that in case the KF
becomes stable, only Σ acc,k in the innovation covariance changes due to external acceleration.
To detect anomalous measurements at time k, Ĉk can be approximated by averaging the
values of Ck over multiple instances of the filtering update, which can be estimated in
real-time by
k
1
Ĉk = ∑ eeT
µ j = k − µ +1 j j
(28)
Sensors 2024, 24, 4623 6 of 16

where µ is the time window length. In order to estimate Σ̂ acc , we set

k
1
∑ e e T = Hk P−
µ j = k − µ +1 j j
T
k Hk + ST + Σ aa (29)

where ST is the matrix variable corresponding to the RKF, then we have

k
1
ST = ∑ e e T − Hk P−
µ j = k − µ +1 j j
T
k Hk − Σ aa (30)

When the condition of Equation (27) is satisfied, the RKF uses Equation (30) to update
the adjustment matrix ST. Due to the limited correlation among the effects of external
acceleration to the vehicle in the three axial directions of the accelerometer, Σ̂ acc,k is updated
as follows:
Σ̂ acc,k = diag(s1 , s2 , s3 ) (31)
si = max (0, Sii ) (32)
where Sii is the ith diagonal element of ST, i = 1, 2, 3, and we set si = max (0, Sii ) to denote
the external acceleration modeling error always larger than 0.
The flowchart of attitude estimation based on multi-scale disturbance compensation
is illustrated in Figure 1. The time window length µ utilized in calculation Ĉk not only
affects Σ̂ acc,k calculation results but also impacts the sensitivity of anomaly detection in
measurements. The selection of µ is a tradeoff among prescribed computational cost,
time-delay and dynamic accuracy of the attitude estimator. The subsequent experiments in
Sensors 2024, 24, x FOR PEER REVIEW 7 of 17
this paper will further explore the effects of varying values of µ on attitude estimation for
low-dynamic vehicles.

Figure
Figure 1.
1. Flowchart
Flowchartof
of attitude
attitude estimation
estimation based
based on
on RKF.
RKF.

3. Results
3.1. Turntable Angle Tracking Test
The effectiveness of the RKF was evaluated through experimentation on a three-axis
continuously rotating turntable, where its performance under static or quasi-static condi-
tions was investigated. The turntable exhibited an inclination rotation error of ±5″, an an-
gular positioning accuracy of ±4″, and a minimum angular rate of 0.001°/s. The attitude
Sensors 2024, 24, 4623 7 of 16

3. Results
3.1. Turntable Angle Tracking Test
The effectiveness of the RKF was evaluated through experimentation on a three-
axis continuously rotating turntable, where its performance under static or quasi-static
conditions was investigated. The turntable exhibited an inclination rotation error of ±5′′ ,
an angular positioning accuracy of ±4′′ , and a minimum angular rate of 0.001◦ /s. The
attitude measurement module is self-designed based on a microcontroller unit (STM32F429)
and MEMS-IMU (KY-IMU102N-A0), where the latter is provided by Beijing Beidou Satellite
Communication Group Co., Ltd. (Beijing, China). The raw IMU data was transmitted
to STM32F429 via a serial peripheral interface, where the constant biases of gyroscope
and accelerometer were well compensated. The gyroscope exhibited a zero-bias stability
of 12◦ /h (with 10 s smoothing), and its zero-bias repeatability was 0.24◦ /h. The zero-
bias stability of the accelerometer was 60 µg, and its zero-bias repeatability was 1 mg.
A widely applied attitude and heading reference system (AHRS), MTi-300, provided by
Xsens Company (Enschede, The Netherlands), was utilized to assess the performance of the
developed algorithm in monitoring the turntable’s angle changes. The MTi-300 consists of
triaxial gyroscopes with zero-bias stability of 10◦ /h, and triaxial accelerometers with zero-
bias stability of 40 µg, which are comparable to our self-designed attitude measurement
board. The nominal attitude accuracy of the MTi-300 is 0.2◦ in root mean square. In Figure 2,
the self-designed attitude measurement board and MTi-300 are affixed to the fixture and
then fixed on the turntable. Initially, the inner axis of the turntable (corresponding to
the Y-axis of the IMU and MTi-300) remained stationary and was subsequently rotated
to the 10◦ position at an angular velocity of 5◦ /s and an angular acceleration of 10◦ /s2 .
The output data from the IMU and AHRS were concurrently recorded at a data sampling
rate of 100 Hz. The turntable data was captured through external synchronized trigger
acquisition, and all the attitude results were synchronized and saved simultaneously for
post-analysis. To guarantee the comparability of the KF and RKF estimation methods, the
Sensors 2024, 24, x FOR PEER REVIEW 8 of 17
same parameters were set for the two filters, with the exception that the RKF was set to
µ = 1.

Figure2.2.Turntable
Figure Turntabletest.
test.

The
The results depicting the
results depicting theestimation
estimationofofthetheturntable
turntabletesttest angles
angles areare presented
presented in
in Fig-
Figures 3 and
ures 3 and 4, where
4, where thethe MTi-300and
MTi-300 andRKF
RKFdemonstrate
demonstratesuperior
superior tracking
tracking of the
the turntable
turntable
attitude
attitudechange
changecompared
comparedto tothe
theKF.
KF.Given
Giventhat
thatthe
thegyroscope
gyroscopeoutput
outputangular
angularrate
rateimpacts
impacts
attitude
attitudeestimation
estimationduring
duringrapidrapidexternal
externalangle
anglechanges
changes(26~28
(26~28s), s),the
theattitude
attitudeestimation
estimation
was
waspredominantly
predominantlyinfluenced
influenced bybythethe
gyroscope’s output
gyroscope’s angular
output angularrate.rate.
In KFIntime updates,
KF time up-
an unmodeled error
dates, an unmodeled errorin Q k −1 in 𝑸
affects the consistency of the predicted covariance matrix
affects the consistency of the predicted covariance esti-
mation. It is noteworthy
matrix estimation. that during these
It is noteworthy angularthese
that during changes, the RKF
angular showed
changes, theaRKF
zero-crossing
showed a
error, which may
zero-crossing result
error, frommay
which the result
significant
from variation of innovation
the significant variationcovariance.
of innovationWhen the
covari-
Rance.
k of the RKFthe
When is increased,
𝑹 of thethe RKF Kalman gain becomes
is increased, small,gain
the Kalman making
becomesthe RKF depend
small, making more
the
RKF depend more on the integration of gyroscope outputs. Once the online computation
of ST decreases 𝑹 , the RKF depends more on accelerometer outputs, resulting in a
changed estimation error in different directions. By taking the turntable output angle as
the reference, Table 1 presents the root mean square error (RMSE) of different methods.
The result indicates that the angle error of the RKF is 0.051°, demonstrating improvements
 attitude estimation during rapid external angle changes (26~28 s), the attitude estimation
was predominantly influenced by the gyroscope’s output angular rate. In KF time up-
dates, an unmodeled error in 𝑸𝑘−1 affects the consistency of the predicted covariance
matrix estimation. It is noteworthy that during these angular changes, the RKF showed a
Sensors 2024, 24, 4623 8 of 16
zero-crossing error, which may result from the significant variation of innovation covari-
ance. When the 𝑹𝑘 of the RKF is increased, the Kalman gain becomes small, making the
RKF depend more on the integration of gyroscope outputs. Once the online computation
on
of the
ST integration
decreases 𝑹 of𝑘 ,gyroscope
the RKF outputs.
depends Once
moretheononline computation
accelerometer of ST resulting
outputs, decreasesinRka,
the
changed estimation error in different directions. By taking the turntable output angle in
RKF depends more on accelerometer outputs, resulting in a changed estimation error as
different directions.
the reference, TableBy taking the
1 presents turntable
the root meanoutput angle
square as the
error reference,
(RMSE) Table 1 methods.
of different presents
the
Theroot mean
result squarethat
indicates error
the (RMSE) of different
angle error methods.
of the RKF The demonstrating
is 0.051°, result indicatesimprovements
that the angle
error of the RKF is 0.051 ◦ , demonstrating improvements of 65.5% and 29.2% compared to
of 65.5% and 29.2% compared to the KF and MTi-300, respectively.
the KF and MTi-300, respectively.

Sensors 2024, 24, x FOR PEER REVIEW 9 of 17

Figure3.3.Results
Figure Resultsof
ofpitch
pitchtracking
trackingtest.
test.

Figure
Figure4.4.Pitch
Pitchestimation
estimationof
ofdifferent
differentmethods.
methods.

Table
Table 1.
1. Pitch
Pitch error
error of
of different
different methods.
methods.
Methods Pitch RMSE (° )
Methods Pitch RMSE (◦ )
KF 0.148
KF 0.148
MTi-300
MTi-300
0.072
0.072
RKF
RKF 0.051
0.051

3.2.
3.2.Dynamic
DynamicAttitude
AttitudeEstimation
EstimationTestTest
Low-dynamic vehicles frequently
Low-dynamic vehicles frequently suffer sufferfrom
from angular
angular or or linear
linear vibrations,
vibrations, whichwhich
may
may degrade attitude estimation based on a MEMS-IMU without
degrade attitude estimation based on a MEMS-IMU without acceleration compensation.acceleration compensa-
tion. In order
In order to further
to further verify
verify thethe performance
performance ofofthe
theRKF
RKFunder
underthethe condition
condition of
of external
external
acceleration disturbance, a dynamic attitude test platform was built based
acceleration disturbance, a dynamic attitude test platform was built based on the automatic on the auto-
matic vehicle shown in Figure 5. The self-designed attitude measurement
vehicle shown in Figure 5. The self-designed attitude measurement board and MTi-300 board and MTi-
300 were fixed on the test vehicle, and the sampling frequency of the IMU
were fixed on the test vehicle, and the sampling frequency of the IMU and MTi-300 were set and MTi-300
were
to 100setHz.
to 100 Hz.test,
In the In the
thetest, the automatic
automatic vehicle vehicle was controlled
was controlled to pass to pass through
through speed
speed bumps
bumps at different
at different speedsspeeds to simulate
to simulate the interference
the interference of external
of external vibration
vibration on theonspecific
the specific
force
force observation
observation of theofaccelerometer.
the accelerometer. To match
To match the dynamic
the dynamic change change
processprocess
of the of the car-
carrier, the
rier,
RKFthewas RKFsetwas
to µset
= 1,toand
μ = the
1, and the automatic
automatic vehiclevehicle
ran at ran at an average
an average speedspeed
of 0.5of
m/s0.5 (low
m/s
(low speed)
speed) andm/s
and 1.5 1.5 m/s (medium
(medium speed),
speed), respectively.
respectively.
 300 were fixed on the test vehicle, and the sampling frequency of the IMU and MTi-300
were set to 100 Hz. In the test, the automatic vehicle was controlled to pass through speed
bumps at different speeds to simulate the interference of external vibration on the specific
force observation of the accelerometer. To match the dynamic change process of the car-
Sensors 2024, 24, 4623 9 of 16
rier, the RKF was set to μ = 1, and the automatic vehicle ran at an average speed of 0.5 m/s
(low speed) and 1.5 m/s (medium speed), respectively.

Sensors 2024, 24, x FOR PEER REVIEW 10 of 17

speed increases, even slight vibrations during the autonomous vehicle’s movement can
significantly
Figure impact
Figure5.5.Attitude
Attitude the attitude
estimation
estimation estimation
testof
test ofthe results.
theautomatic
automatic Taking the attitude of the MTi-300 as
vehicle.
vehicle.
the reference value, the attitude estimation error of the KF and the RKF are shown in Table
The
2. It isThe corresponding
notable that the RKF
corresponding attitude
suppresses
attitude results are
theare
results shown
external
shown inFigures
Figures6interference
acceleration
in 6and
and7.7.Since
Since theattitude
better
the attitude
than
estimation
the KF underof
estimation of the KF speed
different is
is affected
affected by
byexternal
conditions, external acceleration
and its pitch and roll
acceleration interference,
are improved
interference, andanditsabout
by pitch
its hashas
57%
pitch ob-
and 74.8%
vious at
abruptlow speed,
changes, and
the about
RKF can47.4% and 45.7%
effectively at
smooth medium
the abrupt speed.
obvious abrupt changes, the RKF can effectively smooth the abrupt change of pitch by change of pitch by using
innovation
using variance
innovation and external
variance and externalacceleration errorerror
acceleration compensation.
compensation. Comparing
Comparing the atti-
the
Table
tude2.estimation
attitude Attitude error
estimation of different
results
results for
forlowmethods
low speedatand
speed different
and mediumvehicle
medium speeds.
speed,
speed, itit is
is evident that
that as
as the
thevehicle
vehicle
speedMethod
increases, evenAverage
slight vibrations
Speed (m/s)during Pitch
the autonomous
RMSE (°) vehicle’s movement
Roll RMSE (°) can
significantly impact the attitude estimation results. Taking the attitude of the MTi-300 as the
0.5 0.728 0.965
reference KFvalue, the attitude estimation error of the KF and the RKF are shown in Table 2. It
1.5 0.875 0.912
is notable that the RKF suppresses the external acceleration interference better than the KF
under different
RKF speed conditions, and its pitch and roll are improved by0.243
0.5 0.313 about 57% and
74.8% at low speed, and about 1.5 47.4% and 45.7% at medium0.460 speed. 0.495

(a)

(b)
Attitudeestimation
Figure6.6.Attitude
Figure estimation
atat low
low speed
speed with𝜇 µ= =
with 1. 1.
(a)(a) Pitch;
Pitch; (b)(b) roll.
roll.
Sensors
Sensors2024,
2024,24,
24,x4623
FOR PEER REVIEW 11 of 17
10 of 16

(a)

(b)
Attitudeestimation
Figure7.7.Attitude
Figure estimationatatmedium
mediumspeed
speedwith =1.1.(a)
with 𝜇µ = (a)Pitch;
Pitch;(b)
(b)roll.
roll.

Table 2. Attitude error of different methods at different vehicle speeds.

From Figures 6 and 7, it can be seen that the ST values fluctuate significantly when
the vehicle passes over speed
Method bumps.
Average SpeedIn Figure 6,Pitch
(m/s) the RKF
RMSE (◦ ) not show
does RMSE (◦ ) im-
significant
Roll
provement over the KF in the pitch 0.5 between 10 s and 12 s, possibly because the
0.728 condition
0.965
of EquationKF(27) is not met, and the1.5 RKF does not compensate0.875 for the acceleration
0.912 model-
ing error. Notice that, by employing 0.5
μ = 1, the ST values at medium speed clearly
0.313 0.243
increase
comparedRKF with that of the low-speed
1.5 scenario, which coincides
0.460 with the fact0.495 the ex-
that
ternal acceleration of the former scenario is larger. It is notable that the improvement of
roll is better than pitch when the vehicle runs at low speed, while it is not the case when
From Figures 6 and 7, it can be seen that the ST values fluctuate significantly when
the vehicle runs at medium speed, i.e., both of them show similar improvement. The dif-
the vehicle passes over speed bumps. In Figure 6, the RKF does not show significant
ferent performance in roll and pitch estimation indicate that the multi-scale acceleration
improvement over the KF in the pitch between 10 s and 12 s, possibly because the condition
compensation based on Equation (31) can decouple the external acceleration projected in
of Equation (27) is not met, and the RKF does not compensate for the acceleration modeling
different sensitive axes of the IMU. In order to further analyze the influence of Equation
error. Notice that, by employing µ = 1, the ST values at medium speed clearly increase
(27) on attitude anti-interference estimation, μ = 10 was selected to process the IMU data
compared with that of the low-speed scenario, which coincides with the fact that the
at low speed, and the results are shown in Figure 8. Notice that the RKF pitch estimation
external acceleration of the former scenario is larger. It is notable that the improvement
results are improved between 10 s and 12 s by increasing the time window length of the
of roll is better than pitch when the vehicle runs at low speed, while it is not the case
innovation estimation.
when the vehicle Bymedium
runs at increasing the i.e.,
speed, value of of
both μ, them
the innovation
show similarvariance can be esti-
improvement. The
mated more accurately, and thus the sensitivity and precision of the external acceleration
different performance in roll and pitch estimation indicate that the multi-scale acceleration
model are improved.
compensation based on Equation (31) can decouple the external acceleration projected in
different sensitive axes of the IMU. In order to further analyze the influence of Equation (27)
on attitude anti-interference estimation, µ = 10 was selected to process the IMU data at
low speed, and the results are shown in Figure 8. Notice that the RKF pitch estimation
results are improved between 10 s and 12 s by increasing the time window length of
the innovation estimation. By increasing the value of µ, the innovation variance can be
estimated more accurately, and thus the sensitivity and precision of the external acceleration
model are improved.
Sensors 2024, 24, x FOR PEER REVIEW 12 of 17
Sensors 2024, 24, x FOR PEER REVIEW 12 of 17
Sensors 2024, 24, 4623 11 of 16

(a)
(a)

(b)
(b)
Figure 8. Attitude estimation at low speed with 𝜇 = 10. (a) Pitch; (b) roll.
Attitudeestimation
Figure8.8.Attitude
Figure estimationatatlow
lowspeed with𝜇µ==10.
speedwith 10.(a)
(a)Pitch;
Pitch;(b)
(b)roll.
roll.

The
Theattitude
attitude estimation resultswith
estimation results withµ μ = 10
= 10 underunder medium
medium speed speed conditions
conditions are
are shown
shown Thein attitude
Figure 9, estimation
and compared results
with with
Figureμ = it
7, 10 under
can be medium
seen that thespeed
pitch conditions
is are
significantly
in Figure 9, and compared with Figure 7, it can be seen that the pitch is significantly
shown in Figure
improved 9, and compared s. with Figure ST7,value
it can in
be seen9 that the pitch is changes
significantly
improvedininthe therange
rangeof of6 s–8
6 s–8 When
s. When the the ST valuethe s–11
in the s interval
9 s–11 sig-
s interval changes
improved
nificantly, in the
the the range
pitch of
andand 6 s–8
rollroll s.
of the When
RKF the ST
areare value
smoother in the 9 s–11 s interval changes sig-
significantly, pitch of the RKF smootherthan thanthose
thoseofofthetheKF,
KF, and
and maintain
maintain a
nificantly,
a similar
similartrendthe pitch
trendwith and
withthat roll
thatofoftheof the
theMTi-300.RKF are
MTi-300.Comparing smoother
Comparing thethan those of the KF, and maintain
the STST values
values under
under different
different μ selec-
µ selections,
a similar
tions, it trend
can be withthat
seen thatincreasing
of the MTi-300.
μ can Comparing
obtain a moretheaccurate
ST values under different
covariance of the μ selec-
external
it can be seen that increasing µ can obtain a more accurate covariance of the external
tions, it can
acceleration be seen that increasing μ can obtain a more accurate covariance of the external
accelerationmodeling
modelingerror,error,but
butitsitsdynamic
dynamicresponse
responseability
abilitytotoexternal
externalinterference
interferenceisis
acceleration
degraded, modeling error, but itsby dynamic response abilitytrend
to external interference is
degraded, which is further verified by the slow convergence trend of the rollofofthe
which is further verified the slow convergence of the roll theRKF
RKF
degraded,
between which is further verified by the slow convergence trend of the roll of the RKF
between1010s–12s–12s sininFigure
Figure9.9.
between 10 s–12 s in Figure 9.

(a)
(a)
Figure 9. Cont.
 Sensors 2024, 24, x FOR PEER REVIEW 13 of 17
Sensors 2024, 24, x FOR PEER REVIEW 13 of 17
Sensors 2024, 24, 4623 12 of 16

40 800
RKF
KF
MTi-300
30 ST,µ=10 600

20 400

10 200

0 0

−10 −200
0 2 4 6 8 10 12
Time / s
(b)
(b)
Figure
Figure9.9. Attitude estimation
estimationatat medium speed with 𝜇= 10. (a) Pitch; (b) roll.
Figure Attitudeestimation
9.Attitude atmedium
mediumspeed
speedwith
with𝜇µ==10.
10.(a)
(a)Pitch;
Pitch;(b)
(b)roll.
roll.

3.3. Plough
3.3. PloughAttitude
Attitude Estimation
Estimation Test
Test
3.3. Plough Attitude Estimation Test
In this
In this section,
section, thethe result
result of
of plough
plough attitude
attitude estimation
estimation is is reported
reported by by employing
employing aa
In this section, the result of plough attitude estimation is reported by employing a
tractor tillage
tractor tillage platform
platform under
under actual
actual field
field conditions,
conditions, where
where the the uneven
uneven soil
soil conditions
conditions
tractor tillage platform under actual field conditions, where the uneven soil conditions
and soil moisture lead to frequent external acceleration.
and soil moisture lead to frequent external acceleration. As shown in Figure 10, theAs shown in Figure 10,plat-
the
and soil moisture lead to frequent external acceleration. As shown in Figure 10, the plat-
platform is constructed using the Dongfanghong 1104 tractor, with
form is constructed using the Dongfanghong 1104 tractor, with a self-designed attitude a self-designed attitude
form is constructed using the Dongfanghong 1104 tractor, with a self-designed attitude
measurementboard
measurement boardand andanan MTi-300
MTi-300 mounted
mounted on on the the suspended
suspended ploughing
ploughing tool. tool.
To com- To
measurement board and an MTi-300 mounted on the suspended ploughing tool. To com-
compensate
pensate for external
for external acceleration
acceleration of different
of different frequencies,
frequencies, the sampling
the sampling rate
rate of theofIMU
the IMU
and
pensate thefor
andMTi-300 external
MTi-300 areacceleration of different frequencies, thestandard
samplingploughing
rate of theoperational
IMU and
the are set toset
100to 100The
Hz. Hz. The tractor
tractor works works at its
at its standard ploughing operational speed.
the MTi-300
speed. are
Vibration set to 100 Hz.
interference The tractor
comes fromworks
both at its
soil standard
and ploughing
mechanical operational
system during speed.
the work
Vibration interference comes from both soil and mechanical system during the work process
Vibration
process interference
and has a comes from
significant bothon
effect soil and
the mechanical
estimation of system
pitch duringleading
angle, the work to process
obvious
and has a significant effect on the estimation of pitch angle, leading to obvious differences
and has a significant
differences in pitchwitheffect on the
compared estimation of MTi-300.
pitch angle, leading to obvious differences
in pitch compared that ofwith that of the
the MTi-300. The novel KF The novel KF
innovation innovation variance
variance estimation
inestimation
pitch compared with that of the MTi-300. The novel KF innovation variance estimation
method is employed to compensate for the external acceleration. The values of μ = 1 and of
method is employed to compensate for the external acceleration. The values μ
method
µ = 1 is employed
and µ = 10 were to selected
compensate to for the external
evaluate the acceleration.
effectiveness of The
the values of μ = 1 and μ
RKF.
= 10 were selected to evaluate the effectiveness of the RKF.
= 10 were selected to evaluate the effectiveness of the RKF.

(a) (b)
(a) (b)
Figure 10. Attitude
Attitudeestimation
estimationof plough. (a)(a)
Setup forfor
plough attitude estimation; (b) sensors in-
Figure10.
Figure 10.Attitude estimation of of plough.
plough. Setup
(a) Setup plough
for plough attitude
attitude estimation;
estimation; (b) sensors
(b) sensors in-
stallation diagram.
installation diagram.
stallation diagram.

The plough
Theplough pitch
ploughpitch results
pitchresults from
resultsfrom different
fromdifferent methods
differentmethods
methodsand and the
andthe values
thevalues
valuesof of ST
ofST for
STfor different
differentμμµ
fordifferent
The
settings
settings are shown
shown in inFigures
Figures11 11and
and12.12.ItItisis notable
notable that
that the the
RKFRKF mitigates
mitigates pitchpitch varia-
variations
settings are shown in Figures 11 and 12. It is notable that the RKF mitigates pitch varia-
tions effectively
effectively and and exhibits
exhibits a a comparable
comparable result result
to thattoofthat
the of the MTi-300.
MTi-300. UtilizingUtilizing
MTi-300 MTi-300
output
tions effectively and exhibits a comparable result to that of the MTi-300. Utilizing MTi-300
output as the reference
as the reference value, value,
Table 3Table 3 details
details the pitch the pitch estimations
estimations achieved
achieved through
through dif-
different
output as the reference value, Table 3 details the pitch estimations achieved through dif-
ferent methods.
methods. ComparedCompared
with thewith
KF,the
theKF,
RKFthe RKF demonstrates
demonstrates superiorsuperior
suppression suppression of
of external
ferent methods. Compared with the KF, the RKF demonstrates superior suppression of
acceleration
external regardlessregardless
acceleration of time window
of timelength.
window Thelength.
values Theof STvalues
with aof larger time awindow
ST with larger
external acceleration regardless of time window length. The values of ST with a larger
length
time are smoother,
window length which can compensate
are smoother, which can forcompensate
external acceleration
for externalcontinuously
acceleration when
con-it
time window length are smoother, which can compensate for external acceleration con-
occurs during
tinuously whena itshort period.
occurs duringTheapitch
shortresult
period. of The
the RKF
pitchwith µ =of10the
result is closer
RKF with to that
μ =of10the
is
tinuously when it occurs during a short period. The pitch result of the RKF with μ = 10 is
MTi-300
closer compared
to that of the with that compared
MTi-300 of the RKFwithwith thatµ = 1.
ofMore
the RKFspecifically,
with μ =the pitch specifically,
1. More is improved
closer to that of the MTi-300 compared with that of the RKF with μ = 1. More specifically,
approximately
the by 47.5%
pitch is improved and 31.8%, respectively,
approximately by 47.5% and compared with that of compared
31.8%, respectively, the KF. Thewith test
the pitch is improved approximately by 47.5% and 31.8%, respectively, compared with
Sensors 2024, 24, x FOR PEER REVIEW 14 of 17

Sensors 2024, 24, 4623 13 of 16

that of the KF. The test results indicate that external vibration interference during tractor
ploughing operations has a significant effect on the attitude estimation of the plough im-
results indicate that external vibration interference during tractor ploughing operations
plement. The RKF proves to be effective in compensating external acceleration interfer-
has a significant effect on the attitude estimation of the plough implement. The RKF proves
ence.
to be effective in compensating external acceleration interference.

12.8
KF
RKF µ=1
12.7 MTi-300

12.6
Pitch / deg

12.5

12.4

12.3

12.2
0 2 4 6 8 10 12 14 16 18 20
Time / s
(a)

(b)
Figure11.
Figure Ploughpitch
11.Plough pitchofofdifferent
differentmethods,
methods,using
usingRKF
RKFwith
withμµ==1.1.(a)(a)Pitch;
Pitch;(b)
(b)value
valueofofST.
ST.

Table 3. Plough pitch error of different filters.

Methods Pitch RMSE (◦ )

KF 0.493
RKF µ = 1 0.336
Pitch / deg

RKF µ = 10 0.259

To evaluate the effectiveness of Equation (27) on the detection of external disturbances,

this study compares and scrutinizes the attributes of ST value estimation associated with
varying values of µ. When µ is set to 1, ST value exhibits frequent fluctuations, enhanc-
ing its ability to differentiate external acceleration changes and increasing its sensitivity.
Conversely, when µ is set to 10, ST value demonstrates smoother changes, aligning the
pitch angle’s variation more closely with the output of the MTi-300. Therefore, ST values
(a)
under different conditions of µ indicate that higher values of µ lead to a more accurate
estimation of the uncertainty corresponding to unmodeled external acceleration. However,
this results in reduced dynamic responsiveness to changes in actual attitude. Given that
the adjustment of µ necessitates the consideration of carrier attitude response, interference
signal characteristics, and algorithmic computation, future research will focus on dynamic
window length selection for the RKF for vehicles of different dynamic conditions.
 Sensors 2024, 24, 4623
(b)
14 of 16
Figure 11. Plough pitch of different methods, using RKF with μ = 1. (a) Pitch; (b) value of ST.

Pitch / deg

Sensors 2024, 24, x FOR PEER REVIEW 15 of 17

(a)

(b)
Figure12.
Figure Ploughpitch
12.Plough pitchofofdifferent
differentmethods,
methods,using
usingRKF
RKFwith
withμµ= =10.
10.(a)(a)Pitch;
Pitch;(b)
(b)value
valueofofST.
ST.

4. Conclusions
Table 3. Plough pitch error of different filters.
In order to improve attitude estimation of vehicles operating in complex vibration
Methods
conditions, this study introduces a multi-scale compensation Pitch RMSE (°)
approach, termed robust KF
KF
(RKF), for external acceleration of the MEMS-IMU employed in0.493 low-dynamic conditions.
The performance RKFof the
μ = 1RKF was validated through experiments 0.336based on a three-axis
turntable, automatic control
RKF μ = 10 vehicle and tractor tillage test, in which its response to dif-
0.259
ferent vibration disturbances were analyzed. The results of the turntable experiments
To evaluate
demonstrate thatthe effectiveness
the angular trackingof Equation
error of(27)
theon theoutperforms
RKF detection of that external
of the disturb-
KF and
ances, this study
the MTi-300, andcompares
the root andmean scrutinizes
square errorthe attributes
(RMSE) isof ST value from
decreased estimation ◦ and 0.072◦ ,
0.148 associated
with varying values
respectively, to 0.051 of◦ .μ.The
When μ is set vehicle
automatic to 1, STand
value exhibits
tractor fieldfrequent fluctuations,
test demonstrate thaten-
the
hancing its ability
RKF enhances theto differentiate
external external
acceleration accelerationeffect
compensation changes and increasing
significantly its sensi-to
in comparison
tivity. Conversely,
the KF. Furthermore, when theμestimation
is set to 10,ofSTthe
value demonstrates
external accelerationsmoother
modeling changes,
error isaligning
affected
bypitch
the the selected
angle’s window
variationlength
more for innovation
closely with thevariance
output estimation.
of the MTi-300. Decreasing
Therefore, the ST
window
val-
length can enhance real-time attitude estimation, while increasing it can
ues under different conditions of μ indicate that higher values of μ lead to a more accurate boost the accuracy
of externalofacceleration
estimation the uncertainty compensation and thus
corresponding to can accurately
unmodeled estimate
external the attitudesHow-
acceleration. of the
automatic
ever, vehicle
this results in and
reducedplough.
dynamic responsiveness to changes in actual attitude. Given
that theThe experimental
adjustment of μresults presented
necessitates in this paper can
the consideration serve as
of carrier a valuable
attitude reference
response, inter-for
the attitude
ference signalestimation of low-dynamic
characteristics, and algorithmicvehicles, such as monitoring
computation, tractorwill
future research tillage
focusdepth
on
based onwindow
dynamic plough length
attitude estimation.
selection for theFuture
RKF research willof
for vehicles focus on exploring
different dynamicadaptive
condi-
methods
tions. for external acceleration modeling, particularly for tillage depth monitoring in the
presence of different vibration disturbances.
4. Conclusions
In order to improve attitude estimation of vehicles operating in complex vibration
conditions, this study introduces a multi-scale compensation approach, termed robust KF
(RKF), for external acceleration of the MEMS-IMU employed in low-dynamic conditions.
The performance of the RKF was validated through experiments based on a three-axis
turntable, automatic control vehicle and tractor tillage test, in which its response to differ-
ent vibration disturbances were analyzed. The results of the turntable experiments
demonstrate that the angular tracking error of the RKF outperforms that of the KF and the
MTi-300, and the root mean square error (RMSE) is decreased from 0.148° and 0.072°, re-
spectively, to 0.051°. The automatic vehicle and tractor field test demonstrate that the RKF
Sensors 2024, 24, 4623 15 of 16

Author Contributions: Conceptualization, B.C. and J.C.; methodology, B.C. and Y.Z.; software, B.C.
and J.C.; validation, J.C., Y.Z. and Z.S.; formal analysis, Z.S.; investigation, B.C.; resources, Y.L.; data
curation, Y.Z.; writing—original draft preparation, J.C. and B.C.; writing—review and editing, B.C.
and X.W.; visualization, Y.Z.; supervision, B.C.; project administration, X.W.; funding acquisition, B.C.
All authors have read and agreed to the published version of the manuscript.
Funding: This research was jointly funded by the National Natural Science Foundation of China,
grant number 32271999, the Primary Research & Development Plan of Jiangsu Province, grant number
BE2021313, the Jiangsu Province and Education Ministry Co-sponsored Synergistic Innovation Center
of Modern Agricultural Equipment, grant number XTCX2009, the Primary Research & Development
Plan of Danyang City (Modern Agricultural), grant number SNY202303, the Zhejiang Province
Agricultural Machinery Research, Manufacturing and Application Integration Project, grant number
YF20220801, and the Open Funding from the Key Laboratory of Modern Agricultural Equipment and
Technology (Jiangsu University), Ministry of Education, grant number MAET202301. The APC was
funded by XTCX2009.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Tang, C. Non-initial attitude error analysis of SINS by error closed-loop and energy-level constructing. Aerosp. Sci. Technol. 2020,
106, 106149. [CrossRef]
2. Xia, J.; Li, D.; Liu, G.; Cheng, J.; Zheng, K.; Luo, C. Design and Test of Electro-hydraulic Monitoring Device for Hitch Tillage
Depth Based on Measurement of Tractor Pitch Angle. Trans. Chin. Soc. Agric. Mach. 2021, 52, 386–395.
3. Zhang, Q.; Hu, Y.; Li, S.; Zhang, T.; Niu, X. Mounting Parameter Estimation from Velocity Vector Observations for Land Vehicle
Navigation. IEEE Trans. Ind. Electron. 2022, 69, 4234–4244. [CrossRef]
4. Jwu-Sheng, H.; Kuan-Chun, S. A Robust Orientation Estimation Algorithm Using MARG Sensors. IEEE Trans. Instrum. Meas.
2015, 64, 815–822. [CrossRef]
5. Zhang, X.; He, J.; Hua, X.; Chen, Z.; Feng, Z.; Taciroglu, E. Simultaneous Identification of Time-Varying Parameters and External
Loads Based on Extended Kalman Filter: Approach and Validation. Struct. Control Health Monit. 2023, 2023, 8379183. [CrossRef]
6. Shaaban, G.; Fourati, H.; Kibangou, A.; Prieur, C. MARG Sensor-Based Attitude Estimation on SO3 Under Unknown External
Acceleration. IEEE Control Syst. Lett. 2023, 7, 3795–3800. [CrossRef]
7. Chang, J.; Cieslak, J.; Davila, J.; Zhou, J.; Zolghadri, A.; Guo, Z. A Two-Step Approach for an Enhanced Quadrotor Attitude
Estimation via IMU Data. IEEE Trans. Control Syst. Technol. 2018, 26, 1140–1148. [CrossRef]
8. Lee, J.K.; Park, E.J.; Robinovitch, S.N. Estimation of Attitude and External Acceleration Using Inertial Sensor Measurement
During Various Dynamic Conditions. IEEE Trans. Instrum. Meas. 2012, 61, 2262–2273. [CrossRef] [PubMed]
9. Mahony, R.; Hamel, T.; Pflimlin, J.-M. Nonlinear Complementary Filters on the Special Orthogonal Group. IEEE Trans. Automat.
Contr. 2008, 53, 1203–1218. [CrossRef]
10. Madgwick, S.O.; Harrison, A.J.; Vaidyanathan, A. Estimation of IMU and MARG orientation using a gradient descent algorithm.
IEEE Int. Conf. Rehabil. Robot. 2011, 2011, 5975346. [CrossRef]
11. Wilson, S.; Eberle, H.; Hayashi, Y.; Madgwick, S.O.H.; McGregor, A.; Jing, X.; Vaidyanathan, R. Formulation of a new gradient
descent MARG orientation algorithm: Case study on robot teleoperation. Mech. Syst. Signal Process 2019, 130, 183–200. [CrossRef]
12. Farahan, S.B.; Machado, J.J.M.; de Almeida, F.G.; Tavares, J. 9-DOF IMU-Based Attitude and Heading Estimation Using an
Extended Kalman Filter with Bias Consideration. Sensors 2022, 22, 3416. [CrossRef] [PubMed]
13. Rong, H.; Peng, C.; Chen, Y.; Lv, J.; Zou, L. An EKF-Based Attitude Estimator for Eliminating the Effect of Magnetometer
Measurements on Pitch and Roll Angles. IEEE Trans. Instrum. Meas. 2023, 72, 1–10. [CrossRef]
14. Akbari, A.; Rahemi, F.; Khosrowjerdi, M.J.; Ebadollahi, S. Roll and Pitch Estimation from IMU Data Using an LPV H∞ Filter. IEEE
Trans. Instrum. Meas. 2023, 72, 3301864. [CrossRef]
15. Cui, B.; Wei, X.; Chen, X.; Wang, A. Improved high-degree cubature Kalman filter based on resampling-free sigma-point update
framework and its application for inertial navigation system-based integrated navigation. Aerosp. Sci. Technol. 2021, 117, 106905.
[CrossRef]
16. Liu, J.; Liu, H.; Wang, J.; Gu, H. Coordinated Lateral Stability Control of Autonomous Vehicles Based on State Estimation and
Path Tracking. Machines 2023, 11, 328. [CrossRef]
17. Bian, Y.; Yang, Z.; Sun, X.; Wang, X. Speed Sensorless Control of a Bearingless Induction Motor Based on Modified Robust Kalman
Filter. J. Electr. Eng. Technol. 2023, 19, 1179–1190. [CrossRef]
Sensors 2024, 24, 4623 16 of 16

18. Liu, J.; Zhang, R.; Hou, S. Inter-vehicle distance estimation considering camera attitude angles based on monocular vision. Proc.
IMechE Part. D-J. Automob. Eng. 2020, 235, 894–900. [CrossRef]
19. Liu, Z.; Cai, Y.; Wang, H.; Chen, L. Surrounding Objects Detection and Tracking for Autonomous Driving Using LiDAR and
Radar Fusion. Chin. J. Mech. Eng. 2021, 34, 117. [CrossRef]
20. Del Rosario, M.B.; Khamis, H.; Ngo, P.; Lovell, N.H.; Redmond, S.J. Computationally Efficient Adaptive Error-State Kalman Filter
for Attitude Estimation. IEEE Sens. J. 2018, 18, 9332–9342. [CrossRef]
21. Zhou, Q.; Li, Z.; Yu, G.; Li, H.; Zhang, N. A novel adaptive Kalman filter for Euler-angle-based MEMS IMU/magnetometer
attitude estimation. Meas. Sci. Technol. 2021, 32, 045104. [CrossRef]
22. Zhu, C.; Cai, S.; Yang, Y.; Xu, W.; Shen, H.; Chu, H. A Combined Method for MEMS Gyroscope Error Compensation Using a
Long Short-Term Memory Network and Kalman Filter in Random Vibration Environments. Sensors 2021, 21, 1181. [CrossRef]
[PubMed]
23. Gao, G.; Gao, B.; Gao, S.; Hu, G.; Zhong, Y. A hypothesis test-constrained robust Kalman filter for INS/GNSS integration with
abnormal measurement. IEEE Trans. Vehi. Tech. 2023, 72, 1662–1673. [CrossRef]
24. Jwo, D.; Chen, Y.; Cho, T.; Biswal, A. A robust GPS navigation filter based on maximum correntropy criterion with adaptive
kernel bandwidth. Sensors 2023, 23, 9386. [CrossRef] [PubMed]
25. Javed, M.A.; Tahir, M.; Ali, K. Cascaded Kalman Filtering-Based Attitude and Gyro Bias Estimation With Efficient Compensation
of External Accelerations. IEEE Access 2020, 8, 50022–50035. [CrossRef]
26. Odry, A.; Kecskes, I.; Sarcevic, P.; Vizvari, Z.; Toth, A.; Odry, P. A Novel Fuzzy-Adaptive Extended Kalman Filter for Real-Time
Attitude Estimation of Mobile Robots. Sensors 2020, 20, 803. [CrossRef] [PubMed]
27. Li, Y.; Guo, Y.; Gong, L.; Liu, C. Harvesting Route Detection and Crop Height Estimation Methods for Lodged Farmland Based on
AdaBoost. Agriculture 2023, 13, 1700. [CrossRef]
28. He, J.; Luo, X.; Zhang, Z.; Wang, P.; He, J.; Yue, B.; Ding, F.; Zhu, Q. Positioning correction method for rice transplanters based on
the attitude of the implement. Comput. Electron. Agric. 2020, 176, 105598. [CrossRef]
29. Huang, P.; Zhang, Z.; Luo, X.; Liu, Z.; Wang, H.; Lin, Z.; Gao, W.J. Design and test of tilt angle measurement system for agricultural
implements. Trans. Chin. Soc. Agric. Eng. 2017, 33, 9–16. [CrossRef]
30. Yu, J.H.; Park, J.K.; Cheon, S.H.; Byeon, S.J.; Lee, J.W. Development of a rolling angle estimation algorithm to improve the
performance of implement leveling-control systems for agricultural tractors. Adv. Mech. Eng. 2022, 14, 16878132221138310.
[CrossRef]
31. Yang, H.; Zhou, J.; Qi, Z. Prediction model of pitch angle of greenhouse electric tractors based on time series analysis. Dyna 2023,
98, 620–626. [CrossRef] [PubMed]
32. Zhao, N.; Zhao, B.; Yi, S.; Zhou, Z.; Che, G. Research on a Sowing Depth Detection System Based on an Improved Adaptive
Kalman Filtering Method. Electronics 2022, 11, 3802. [CrossRef]
33. Candan, B.; Soken, H.E. Robust Attitude Estimation Using IMU-Only Measurements. IEEE Trans. Instrum. Meas. 2021, 70, 3104042.
[CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.