2022 2nd International Conference on Algorithms, High Performance Computing and Artificial Intelligence (AHPCAI)

Research on Trajectory Tracking Control of

2022 2nd International Conference on Algorithms, High Performance Computing and Artificial Intelligence (AHPCAI) | 979-8-3503-3478-4/22/$31.00 ©2022 IEEE | DOI: 10.1109/AHPCAI57455.2022.10087391

Intelligent Vehicle Based on PSO-MPC Algorithm

Ran Li1,2,*, Haibo Dong1,2, Jinwei Guo1,2, Xiaoyan Huang1,2, Ziyan Li1,2, Ning Sun1,2
1
CATARC Intelligent and connected technology Co.,LTd., Tianjin, China
2Automotive Data of China Co.,LTd., Tianjin, China

[liran, donghaibo, guojinwei, huangxiaoyan, liziyan, sunning]@catarc.ac.cn

Abstract. In order to solve the problems that low tracking Some scholars have done a lot of research on model
accuracy and slow response speed in the process of intelligent predictive control at home and abroad. Li Yaohua et al.[9]
vehicle track tracking, the model predictive controller optimized designed an adaptive dual-time domain MPC parameter path
by particle swarm optimization(PSO-MPC) algorithm is tracking controller to achieve accurate tracking of the
proposed. And the performance of the algorithm is improved by reference path. Wang Yin et al.[10] used an adaptive MPC
optimizing the weight matrix of the controller in real-time. The controller to study on the method of tracking target route in
model predictive controller with particle swarm optimization is low-speed environments and low-ground adhesion conditions.
built by Matlab software to test the algorithm effect, and the Li Peiqing et al.[11] proposed an improved MPC algorithm to
model of vehicle dynamic is build in Carsim software. The
automatically adjust the weight matrix of the objective
double-shifting track is used as the target trajectory to test the
trajectory tracking capability of the vehicle. By analyzing the
function based on the trajectory curvature, which solved the
simulation results, it can be concluded that the PSO-MPC impact of the trajectory curvature on the tracking accuracy.
algorithm effectively improves tracking response speed and Yuan Fengxia et al.[12] used the genetic method to optimize
accuracy comparing with the traditional MPC algorithm. the MPC’s weight matrix , which can enhance the control
capability of the MPC controller. Due to the complexity of the
Keywords: Intelligent vehicle; Track tracking; Model predictive ordinary algorithm, the computer has a huge amount of
control; PSO； computational load, and each decision cycle takes a long time.
However, vehicles need real-time high-precision tracking
I. Introduction algorithms. It is difficult to meet the needs of high-precision
With the continuous upgrading of the new automobile's vehicle operation. In order to further reduce the computational
four modernizations, unmanned driving has gradually entered load of the algorithm and improve the real-time tracking of the
people's lives and become a new development trend of target route. The PSO-MPC model proposed in this paper can
automobiles[1][2]. As one of the key technologies of unmanned effectively meet the requirements of vehicle accuracy. The
driving, decision-making control is directly related to vehicle main process is as follows: first, the paper builds a model
safety[3], stability and comfort. How to strengthen the vehicle predictive controller witch optimized by particle swarm
control accuracy and response speed of the lateral control[3], optimization (PSO-MPC) and a vehicle dynamics model to
reduce the vehicle lateral movement deviation, and achieve simulate vehicle actually driving mode. Second, combined
accurate tracking of target trajectory is a hot issue in the field with the joint simulation environment of MATLAB/Simulink
of unmanned driving[4]. and Carsim, a virtual simulation model is built. Finally, the
control capability of the PSO-MPC algorithm is tested and
Vehicle trajectory tracking generally refers to effective verified by taking the double lane changing condition as the
tracking of the predicted target route by controlling steering trajectory tracking test curve.
angle to make certain of the safety of the driving vehicle[5]. At
present, there are three kinds of tracking algorithms[6]: the first II. Establishment of the vehicle dynamics model
method is the tracking control algorithm based on the In order to more accurately simulate the actual state of the
principle of road geometry, which is composed of a pure vehicle, the vehicle dynamics model with 3-TOF is build[7].
tracking algorithm, Stanley control, etc[7][8]; the second But the complexity of the model is very high, it requires a lot
method is the tracking algorithm based on classical control of calculation force to meet model running speed requirements.
theory, which mainly includes PID control, linear feedback it is inappropriate to use that to study the vehicle tracking of
control, etc; the last tracking control method for target route is target route. therefore, the model needs to be simplified.
built on the modern control theory algorithm. such as optimal According to the need for vehicle tracking[13], the model
control, MPC, etc. Among them, the MPC algorithm has makes the some assumptions:
strong robustness, which can effectively deal with the
uncertainty, parallelism and nonlinearity during the driving of (1) Assuming that the target vehicle is a rigid body, and
unmanned vehicles, which can effectively solve the various suspension will not cause roll, pitch and vertical motion of the
constraints of controlled variables and control variables. vehicle;
Therefore, the algorithm is widely used in the field of (2) Assuming that aerodynamics during vehicle driving is
driverless vehicles. not considered.

979-8-3503-3478-4/22/$31.00 ©2022 IEEE 148

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.
 (3) Assuming that the vehicle is driven freely on flat centroid to the front/rear axles, VX represent the x-axes
ground
vehicle speed, VY represent the y-axes vehicle speed.
(4) Assuming that the lateral load of the vehicle tires does
not transfer at all. The tire model is very important in vehicle dynamics,
which directly affects the calculation accuracy of the dynamic
(5) Assuming that only the front wheels of the vehicle are model. Therefore, the GIM tire model is selected to build the
used for steering, the influence of the track width on the relationship between tire force and structural parameters, as
steering radius is ignored. shown in Formula (2).
(6) Assuming that the vehicle yaw angle and heading angle
have the same value. [C S l 2 + μ x Fz (1 − 3ln2 + 2ln3 )]sgn( S s ) S s < S sc
Fx =  s s n Fxf
According to the above assumptions, simplify the  μ x Fz sgn( S s ) S s > S sc (2)
completed three degrees of freedom model, and the schematic [Ca Sa ln2 + μ y Fz (1 − 3ln2 + 2ln3 )]sgn( S s ) S s < S sc
diagram is shown in Figure 1. Fy = 
 μ y Fz sgn( Sa ) S s > S sc

δ In the above formula, Fx and Fy represent the tire

Y F xf
longitudinal/later force, Cs and Ca rerepresent
α
ϕ longitudinal/lateral stiffness of the tire, μ x represent the
F yf
longitudinal adhesion coefficient, μ y represent the lateral
vy vx adhesion coefficient. Ss and S ac represent the absolute
lf longitudinal/lateral slip rate of the vehicle tire, Ss and Sa
F xr
represent absolute tire slip rates of x-axle/y-axle direction, ln
lr is the tire contact length in the ground; Fz is the tire vertical
F yr load; S as and S ac represent longitudinal and lateral slip rates.

O
X State quantity ξ = [v y , vx , ϕ , ϕ , Y , X ]T and control
quantity u = [δ ] are selected to establish the nonlinear
Figure 1. Vehicle model vehicle dynamics model with 3-TOF
system, as shown in Formula (3).
Through mechanical analysis of the model with 3-TOF, a
mechanical model is formed by applying Newton's second law, ξ = f (ξ , u ) (3)
and the formula (1) is obtained. Because the control track has an incomplete representation
of state quantity and control quantity, it is impossible to
max =mv yϕ + 2 Fxf cos δ + sFxr linearize according to the reference track, so it is assumed that
 the deviation between the current state ξ and the next time is
ma y =mvxϕ + 2 Fyf cos δ + sFyr
 linearized. Assuming that the control quantity is unchanged, it
 I zϕ = 2l f ( Fxf sin δ + Fyf cos δ ) − 2lr Fyr (1) is shown in Formula (4).

VX = vx cos ϕ − v y sin ϕ ξnext = f (ξnext , unext ) (4)
V = v sin ϕ − v cos ϕ
 Y x y
In the above formula, ξnext = ξ + ξnext T , T is the
In the above formula, m represents vehicle mass; vx sampling time.
represent x-axle direction speed, v y represent y-axle vehicle Taylor series expansion is carried out for system state ξnext ,
speed. ax and ay represent the longitudinal/lateral vehicle and Formula (5) is obtained by ignoring higher-order terms.
acceleration, ϕ is vehicle yaw; represent the front wheels ∂f (ξ , u )
ξnext = f (ξ next , unext ) + (ξ − ξ next )
longitudinal force, Fyf and Fyr represent the lateral force ∂ξ
(5)
exerted by the ground on the front/rear wheel, Fxr and Fxr ∂f (ξ , u )
+ (u − unext )
represent the longitudinal force exerted by the ground on the ∂μ
front/rear wheel. I z represent the moment of vehicle inertia;
According to Formulas (4) and (5):
l f and lr represent the linear distance from the vehicle body
ξ = A(t )u (t ) + B (t )ξ (t ) (6)

149

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.
 ∂f ∂f  η (t + 1| t )   
 CA
Where, A(t ) = ξt ut , B (t ) = ξt ut .  η (t + 2 | t ) 
t ,t
  2 
∂μ ∂ξ   CA
 t ,t 
     
The Euler formula of the preceding term is applied to Where, Y (t ) =   ， Φ t =    Nc  ，
discretize Formula (6), and discrete state space can be η (t + N c | t )   CAt ,t 
calculated, the specific form is shown in Formula (7).      
   
η (t + N p | t )  CA
  Np 
ξ (k + 1) = A(k )u (k ) + B(k )ξ (k ) (7)  t ,t 
Θt =
Where, T is the sampling step size in discretization, as  CB  0  0 
shown in
t ,t
 CA  B 
CB  0 
 t ,t t ,t t ,t

 1 −ϕT 0 −v x T 0 0      ，
 ϕ
T 1 0 v yT 0 0    N  
 , CA Bt , t   N −1 B
CA  CB
 
c c
t ,t t ,t t ,t t ,t
 0 0 1 T 0 0  
B(k) = I + TB(t ) =      
 0 0 0 1 0 0  
 T * cos ϕ T *sin ϕ   N B
 CA N
CA
−2
Bt , t  
CA
N − N −1
Bt , t 
1 0
p p p c

κ *T 0 t ,t t ,t t ,t t ,t
 
 −T *sin ϕ T * cos φ −κ * T 0 0 1
 Δu (t | t ) 
 2T   Δu (t + 1| t ) 
 m ( Fxf cos δ − Fyf sin δ )  ΔU (t ) = 
 
.

   
 − 2T F sin δ − F cos δ  Δu (t + N | t )
 m ( xf )  c 
yf

  In order to ensure the accurate tracking of the target

A(k ) = TA(t ) =  0 , vehicle to the reference track, the objective function is
 2l f T  designed based on the deviation of the state quantity and
 ( Fxf cos δ − Fyf sin δ )  control quantity of the target vehicle, such as Formula (10).
 Iz  Np Nc
 0 J (t ) =  η (t + i | t ) − η ref (t + i | t ) +  Δ u (t + i | t )
2 2
+ ρε 2
  Q R
i =1 i =1

 0  (10)
κ =vx cos ϕ − v y sin ϕ . Where, N p is the prediction time domain, Nc is the control

III. Establishment of the prediction model time domain, Q and R are the weight matrix, ρ is the
weight coefficient; ε represents the relaxation
A. Establishment of the objective optimization function factor.B.Design function constraints
The state space of the objective function is established The vehicle comfort is very important in the condition of
according to formula (7), ensuring safe driving[14], The constraint of control quantity is
  Formula (11), the transfer equation of control quantity is
ξ ( k + 1 t ) = A(k t )ξ (k t ) + B( k t )ΔU (k t ) Formula (12).
 (8)
η ( k t ) = C ξ (k t )
umin (t + k ) ≤ u (t + k ) ≤ umax (t + k ) k = 0,1,..., Nc − 1 (11)

A Bk ,t  B  Δumin (t + k ) ≤ Δu (t + k ) ≤ Δumax (t + k ) k = 0,1,..., Nc − 1
Where, A (k t ) =  k ,t ， B (k t ) =  k ,t  ， C is the output
0mxn I m   Im  u (t + k )=u (t + k − 1) + Δu (t + k ) (12)
matrix, n and m represents the number of state quantities and
control quantities respectively. In the above formula, let U t = I Nc ⊗ u ( k − 1) ,
The predictive output function of the control system is
shown in formula (9). 1 0   0
1 1 0  0 
Y (t ) = Φ tξ (t | t ) + Θt ΔU (t ) (9) 
D=1 1 1  0 ⊗ I m , and the following
 
    
 1 1 1  1  N × N
c c
formula is obtained.
ΔU min ≤ DΔU t + U t ≤ ΔU max (13)

150

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.
 In the above formula, I N c is column vector of size N c , I m the tracking effect of vehicles. In Formula (10), Q and R
generally select the diagonal matrix, which is the state error
represents the m × m unit matrix, ⊗ is Kronecker product, matrix, reflecting the vehicle’s tracking ability to the target
and u (k − 1) represents the control quantity at time k-1. track; R = diag (r1 , r2 ,..., rj ), j = N c × m is the control
The expression form of the general solution of quadratic matrix, which represents the constraint on the control quantity.
programming that converts the objective function into a
simple solution, such as Formula (14). The weight coefficient q directly affects the change
amplitude of the control quantity, and the weight coefficient r
J (t ) = [ΔU (t )T , ε ]T H t [ΔU (t )T , ε ] + Gt [ΔU (t )T , ε ] (14) controls the sensitivity of the system response, and
s.t. ΔU min ≤ ΔU t ≤ ΔU max and U min ≤ U t + DΔU t ≤ U max q = q0 + a1 × 10b , r = r0 + a2 × 10b , a1, b1, a2, b2 are adjustment
1 2

parameters, q0 and r0 are the initial value of q and r.

ΘT QΘt 0
In the above formula, Ht = [ t ] , Real time optimization of Q and R matrix parameters of
0 ρ the current model using PSO algorithm. the evaluation
Gt = [2EtT QΘt 0] , Et = Φt ξ is the deviation between function is defined based on the maximum lateral position
vehicle status and reference status. deviation, average lateral position deviation and average yaw
angle deviation of the vehicle. As shown in Formula (16), it is
By applying the above calculation method, the first control used as the comprehensive evaluation index of the weight
quantity of the control sequence can be calculated , and the coefficient.
relevant state of the system can be updated to complete the
state calculation at this moment. The same method is used to f (t ) = λ1 Ed max + λ2 Ed m + λ3 Eϕ m + λ4 β max + λ5ωmax (16)
calculate the state of the new moment and repeat it
continuously to realize the tracking of the target path by the In the above formula, Ed max is the maximum lateral
intelligent vehicle. position deviation, Edm is the average lateral position
IV. PSO algorithm deviation, Eϕ m is the average yaw angle deviation, and λ is
A.The principle of PSO the weight coefficient.
The principle of PSO algorithm: there is a particle swarm
consisting of M particles moving in the space with dimension
Ed max = max[ Ed (i )], i = 1, 2..N
N, and the particle motion speed is constant, if the real-time 1 N
position of one particle represents a potential solution for the Edm =  Ed (i)
N i =1
(17)
target problem[15]. The fitness value for each potential solution
is calculated through the customized fitness function, and this 1 N
is used as the standard to screen each particle[16]. Eϕ m =  [ϕ (i ) − ϕr (i )]
N i =1
X i = [ xi1 , xi 2 ,..., xiN ] represents the No.i particle’s position,
Vi = [vi1 , vi 2 ,..., viN ] represents the particle speed, the best In the above formula, Ed means the lateral deviation from
the unmanned vehicle to the reference path, N is total number,
position of particle in history is Pi = [ pi1 , pi 2 ,..., piN ] , and the
ϕ references the actual vehicle yaw, and ϕr is the reference
best position of particle swarm in history is
vehicle yaw.
Pg = [ pg1 , pg 2 ,..., pgN ] , g ∈{1, 2,..., M } .
In order to improve the driving stability during path
The velocity and position of every particle will be updated tracking and prevent the vehicle from rolling over, the sideslip
in each round of iterative calculation.and the updated equation angle β and yaw speed ω in the centroid are constrained, as
is shown in Formula (15). shown in Formula (18).
vi N (t + 1) = ωviN (t ) + c1r1 (t )[ piN (t ) − xiN (t )] + c2 r2 (t )[ piN (t ) − xiN (t )] (15)
 β max ≤ arctan(0.02 μ g )
 xi N (t + 1) = xi N (t ) + viN (t + 1)
0.85μ g (18)
where: ω references inertia factor, c1 and c2 are learning ωmax ≤
vx
factors, they are independent random numbers distributed on
[0,1]. Assuming f ( x) is the minimized objective function, p In the above formula, μ references the ground adhesion
coefficient, and g references the gravitational acceleration.
represents the position of particle i,
 pi (t ), f ( xi (t +1）≥ f (pi (t )) This paper mainly describes the optimization method of
pi (t + 1) =  . control accuracy and response speed, and the values of λ1 , λ2 ,
 xi (t + 1), f ( xi (t +1）≥ f (pi (t ))
λ3 , λ4 , λ5 are 100, 200, 100, 59 and 10 respectively
B.Optimization algorithm for constructing weight parameters according to the difference of each parameter dimension to
The matrices Q and R of the MPC controller directly affect meet the maximum evaluation of the weight parameters.

151

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.
 V. Experimental test and result analysis
Reference Track Initialization
A.Test condition Parameters
To test the functional effectiveness of the mode, the paper
selects double lane change condition curve as the target curve
for tracking test to test the vehicle tracking stability and
MPC Controller
tracking accuracy of the vehicle under this condition. The
working condition equation of the double traverse is shown in Update Status
Formula (19), and the working condition diagram is shown in
Vehicle Model
Figure 2.
 1
Yref ( X ) = 2 ( d y1[1 + tanh( z1 )] − d y 2 [1 + tanh( z2 )]) Particle Swarm
 (19) Optimization for Q/R
 d y1 d y2
ϕ ref ( X ) = arctan[1.2 * ( − )]
 d x1 cosh( z1 ) 2
d x 2 cosh( z2 ) 2 Update Q/R
In the above z1 = 0.6 * X − 15.11 ,
formula,
z2 = 0.11* X − 4.97 , d x1 =25, d x 2 =21.55, d y1 =5.0, d y 2 Figure 3. Simulation Test Process

=5.0. The basic parameters of the target vehicle are shown in

Table 1, these parameters are the key parameters of the
vehicle model, including curb weight, vehicle size, tire
parameters, etc.

Table 1. Vehicle Basic Parameters

Parameter Name Value

Equipment quality (kg) 1600
Distance between front axle and
1.4
centroid (m)
Distance between centriod and front
1.65
axle (m)
Front wheel longitudinal and
66900
transverse cornering stiffness (N/rad)
Rear wheel longitudinal and transverse
62700
cornering stiffness (N/rad)
Figure 2. Reference track of double traverse Vehicle tire rolling radius (m) 0.32
Wheel slip ratio 0.2

B.Simulation test process and result analysis

The parameters of the model controller are shown in below
To simulate and test the performance of the model, the Table 2, Mainly including time domain parameters, sample
vehicle dynamics model is built in CarSim software, and the parameters, initialization parameters, etc.
process control and algorithm realization are completed in
MATLAB, forming a complete simulation environment. First, Table 2. Model Controller Parameters
initializing the vehicle parameters, model parameters and
algorithm parameters. Second, combining the double lane Parameter Name Value
change condition reference trajectory, controlling the vehicle
Prediction time domain 20
to travel along the target trajectory by the MPC controller, and Control time domain 8
update the basic parameters in the vehicle model in real time. sampling period (s) 0.02
Finally, aim to Q and R parameter matrices, PSO algorithm is Front wheel deflection control amount (°) -10~10
applied for real-time optimization, and the relevant parameters Front wheel deflection control increment (°) -0.847~0.847
in the model are updated at the same time, so as to achieve the Q initial value 100*eye(4,4)
simulation calculation of the model. The algorithm flow is R initial value 10
S 1000
shown in Figure 3. ρ 1000

Besides what the parameters listed in the above table, the

follow variables also need to be initialized: the adhesion
coefficient value of the simulated road surface is 0.2, the
control time domain value of the controller is 8, and the
prediction time domain value is 20. At the same time, speed
variables are added to test the performance of the model at

152

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.
different vehicle speeds such as 20km/h, 40km/h and 60km/h. of the control model are shown in Figure 4, and the heading
Under the condition of two lane shift, the position test results angle deviation is shown in Figure 5.

Figure 4. Comparison of vehicle track tracking Figure 5. Comparison of vehicle yaw rate

Through comparing and analyzing simulation test result, it Industrial Electronics and Applications (ICIEA). IEEE, 2017.
can be concluded that the test model controller can accurately [5] Cao J, Song C, Peng S, et al(2020). Trajectory Tracking Control
track the target track when speed is 20km/h. the robustness of Algorithm for Autonomous Vehicle Considering Cornering
Characteristics. IEEE Access, PP(99):1-1.
the algorithm is be tested by double lane shifting condition at
[6] Morin P, Samson C(2006). Trajectory tracking for non-holonomic
vehicle speeds of 20km/h, 40km/h, and 60km/h. It can be vehicles: Overview and case study. International Workshop on Robot
found that the tracking accuracy of the vehicle will be Motion & Control. IEEE.
decreased with the speed increase, but the range of change [7] Ge Chang, Qian Suqin(2021). An Adaptive MPC Trajectory Tracking
will gradually decrease. The main reason is that the tracking Algorithm For Autonomous Vehicles. 2021 17th Internation
accuracy of the trajectory will be changed with the difference [8] Wang Fukang(2020). Research on control of regenerative braking
in the vehicle speed and the ground adhesion. In general, the system in hybrid electric vehicle based on composite structure Machine,
tracking accuracy and corresponding time of the intelligent Chongqing University. (in Chinese).
vehicle will be effectively improved by real-time dynamic [9] Li Yaohua, Fan Jikang, Liu Yang, et al(2021). Path planning and path
optimization of the weight coefficients Q and R of the tracking control for autonomous vehicle based on MPC with adaptive
dual-horizon-parameters. Journal of Automotive Safety and Energy 012
predictive control model. - 004. (in Chinese).
[10] Wang Yin, Zhang Haoqi, Sun Qianlai, et al(2021). Research on
VI. Conclusion trajectory tracking control based on adaptive MPC algorithm. Computer
The paper proposes a predictive control model with Engineering and Applications 57 (14): 8. (in Chinese).
optimized by the PSO algorithm. Firstly, a vehicle trajectory [11] Li Peiqing, Yang Zhenglong, Chen Yikai, et al(2022). Research on MPC
tracking MPC controller is designed which is build on the trajectory tracking under variable weight matrix. Machine Tools &
Hydraulics 50 (7): 7. (in Chinese).
vehicle dynamics model. Secondly, in order to tracking the
target track accurately and quickly, the weight coefficients Q [12] Yuan Fengxia, Zhang Hua, Chen Feng, et al(2021). Trajectory Tracking
Control of Intelligent Vehicle Based on Model Predictive and Genetic
and R of MPC algorithm are dynamically optimized by using Algorithm. Journal of Anhui University of Technology(Natural Science)
the particle swarm optimization algorithm. Finally, the 38(04) : 393 - 400. (in Chinese).
simulation environment is build by MATLAB and Carsim [13] Gong Jianwei, Jiang Yan, Xu Wei(2014). Model predictive control of
software, and the PSO-MPC algorithm is validated in the self-driving vehicles. Beijing University of Technology Press. (in
condition of double lane shifting working condition. and the Chinese).
result show that the predictive control model optimized by [14] Tao Bingbing, Zhou Haiying, Wang Sishan(2017). Design of LQR
particle swarm optimization algorithm and dynamically trajectory tracking controller for autonomous vehicle. Journal of Hubei
University of Automotive Technology 31 (4): 7. (in Chinese).
adjusted parameters in real-time can achieve more accurate
[15] Yan S, Li X, Zhang S, et al.(2015) A novel statistical method for step-
and faster tracking of the target trajectory. down stress accelerated life tests using PSO and MLE. International
Conference on Reliability. IEEE, 2015.
References
[16] Li Peiqing, Yang Zhenglong, Chen Yikai, et al(2022). Research on MPC
[1] Qian Yubao, Yu Misen, Guo Xutao, et al(2022). Development of trajectory tracking under variable weight matrix. Machine Tools and
intelligent control technology for unmanned vehicle. Science Hydraulics 50 (7): 7. (in Chinese).
Technology and Engineering 22(10), 3846 - 38581. (in Chinese).
[2] Tan Hua, Hong Weiquan, and Zhang Haitao (2021). Development Trend
of Automatic Driving at Home and Abroad. Communication Enterprise
Management 03, 54 - 57. (in Chinese).
[3] Wongpiromsarn T, Topcu U, Murray R M(2013). Synthesis of Control
Protocols for Autonomous Systems. Unmanned Systems 1(01): 21 - 39.
[4] Niu J , Lu Y , Lin C .(2017) Effect of the tire lateral force saturation on
stability control of 4WDEV steering. 2017 12th IEEE Conference on

153

Authorized licensed use limited to: Wojskowa Akademia Techniczna. Downloaded on June 03,2024 at 16:28:08 UTC from IEEE Xplore. Restrictions apply.