See discussions, stats, and author profiles for this publication at: [Link]

net/publication/308808774

A novel data-driven control for fixed-wing UAV path following

Conference Paper · August 2015

DOI: 10.1109/ICInfA.2015.7279812

CITATIONS READS
6 533

5 authors, including:

Shulong Zhao Xiangke Wang

National University of Defense Technology National University of Defense Technology
43 PUBLICATIONS 493 CITATIONS 184 PUBLICATIONS 2,533 CITATIONS

SEE PROFILE SEE PROFILE

Weiwei Kong
University of Hamburg
30 PUBLICATIONS 444 CITATIONS

SEE PROFILE

All content following this page was uploaded by Shulong Zhao on 10 January 2017.

The user has requested enhancement of the downloaded file.

 A Novel Data-driven Control for Fixed-Wing UAV
Path Following
Shulong Zhao, Xiangke Wang, Daibing Zhang, Lincheng Shen
School of Mechanics and Automation
National University of Defense Technology
Changsha, P. R. China
Email: jaymaths@[Link]

Abstract—In this paper, a novel data-driven control algorithm field approach provides asymptotic following for straight-
is proposed to adaptive control for fixed-wing (Unmanned Aerial line and circular paths in the presence of constant wind
Vehicle) UAV path following using input and output (I/O) data. disturbances. A solution to the trajectory-tracking and path-
Differently from model based control, data-driven control reduces
the dependence of accurate mathematics model in process of following problem for underactuated autonomous vehicles in
controller designation. The key difficulty in fixed-wing UAV path the presence of possibly large modeling parametric uncertainty
following is that the model of UAV is generally underactuated and was proposed in [12]. The control strategy is combined with
nonlinear, and easy to be affected by disturbances. The unique a nonlinear Lyapunov-based tracking technique to obtain a
feature of this paper is that we explicitly employ data-driven hybrid controller, which is robustness to parametric modeling
control to design controller directly in two dimensions at constant
height to overcome the drawback of dynamical uncertainties and uncertainty. The problem of path following using a quadrotor
wind disturbances. The tracking error is guaranteed converging equipped with a fully embedded imaging and control system
to 0 and simulation results in MATLAB environment show the was addressed in [13]. A switching between the measure-
effectiveness and applicability of the control law. ments coming from imaging and inertial sensors was used
to estimate the vehicle parameters required in both regions
I. I NTRODUCTION
of operation, and stability of the vehicle was guaranteed
Recently, researches on control of (Unmanned Aerial Vehi- via a switching control strategy. A dynamic robust path-
cle) UAV have attracted great attentions from a great deal of following AUV (autonomous underwater vehicle) control has
research communities [1][2]. Many controllers have been suc- been developed in [14]. The proposed method uses a hybrid
cessfully developed in various kinds fields [3], which include robust scheme, relying on classic adaptation scheme design of
formation control[4], auto landing[5], and path-following[6]. those dynamic parameters that appear with an affine form, and
Among them, path following has been deeply investigated on switching control for the others. In [7], a detailed analysis
in the literature and a basic requirement is that they must of five path-following algorithms was performed with different
be accurate and robust to wind disturbances [7]. For small parameter settings and wind disturbances. The simulation
and low-cost UAV, it is meaningful to consider the UAV results show that VF path-following technique and NLGL
to autonomously follow a predefied path. It can be seen in (nonlinear guidance law) algorithm are more accurately than
many practical applications such as take a picture, surveillance, other strategies. Arogeti proposed a controller, which utilized
investigation, monitor/report wildfire in forests or traffic in a new transformation that, for the pure rolling case, transfers
urban [8]. To extend the usefulness of UAV beyond their the model to a chain form [15]. The transformation avoid
current applications, the capability of path following accurate model singularity and the convergence rate is proportional to
is of great significant. the vehicle speed and independent of the vehicle heading.
It is well known that reason of limiting achievable tracking The common characters of those methods are that kinemati-
precision is input coupling, wind disturbances, dynamical cal models or dynamical model, are widely used for the control
uncertainty, nonlinear model and performance of sensing and system. However, the uncertainties of model and parameters
control [9]. Because of those factors, finding true accurate and were rarely considered. Data-driven control (DDC), using I/O
robust control algorithm in fixed-wing UAV path following is data to design controller online or offline, is more simple
difficult. Meanwhile, designing a controller to deal with such and independent on priori information of model. There are
a complex system is not a simple process, and classic linear more than 10 kinds of DDC methods, and model-free adaptive
approaches omit some dynamical characters necessarily and do control (MFAC) is proposed for general nonlinear systems.
not lead to satisfactory performances [10]. In early methods, In [16], the estimation of pseudo partial derivative (PPD) is
various assumptions or simplifications on the UAV dynamics obtained by gradient descent (GD) algorithm. While, without
were used to obtain near-accurate solutions. any priori acknowledge, it is difficulty to settle initial value of
A method for UAV path following, based on the concept of parameters and there is little guidance for adjustment of pa-
vector fields (VF), has been introduced in [11], and Lyapunov rameters, which is easy lead to local minimum or oscillations
stability criteria are used. it has been shown that the vector in earlier stages even result in fail under some circumstances.
Inspired by [19], we employ radial basis function (RBF) neural
network to estimate PPD. It is practical and valid that iterative
estimation of PPD and resetting algorithm will be omitted.
Contrary to other mentioned approaches, the method pro-
posed in this paper does not require acknowledges of system
model or analysis of dynamical rule of UAV. The most
contributions of this paper are:
1) A novel data-driven control algorithm is proposed to
adaptive control for fixed-wing UAV path following
without employing any acknowledges of model of UAV;
2) The PPD is estimated by RBF neural network, which
is necessary for estimation of PPD when the control
scheme is hybrid.
3) The stability of proposed controller is presented with Fig. 1. Coordinate system definitions of the UAV in flight
discrete Lyapunov technique.
Consequently, the proposed controller is simple and easy
the minimum and maximum of control input, and these are
for implementation. There is no requirement for cost and ex-
considered as constraints in proposed controller.
tremely time consuming process of system identification, and
The key challenge to achieve these objectives is the existing
the bounds of uncertain disturbances are assumed unknown.
unmodeled dynamics and alterable structure of UAV. In addi-
The proposed algorithm, although easy for implementation,
tion, wind disturbances and noises will increase the difficulty
still shows significant performance improvement when random
of accurate control.
disturbances are not negligible.
This paper is organized as follows. the primary introduc- B. Preliminary of MFAC
tion of MFAC method and some necessary assumptions and Consider a class of discrete SISO system
notations are given in Section II. Data-driven path-following
control laws are developed in 2-D at constant height in y(k + 1) = f (y(k), . . . , y(k − ny ), u(k), . . . , u(k − nu )) (3)
section III. The performance of the control system proposed where, y(k) ∈ R, u(k) ∈ R are the output and input of
is illustrated in simulation in Section IV. Section V concludes system at time k, respectively. ny and nu are unknown positive
this paper. constants. f (·) is a general nonlinear function.
Assumption 1: The system (3) is observable and control-
II. I NTRODUCTION
lable.
A. Background and Problem Statement Assumption 2: Omit some finite time point, the partial
In the most general path following problems, the objective derivative of f (·) with respect to control input u(k) is contin-
of developed method is to accurately follow a a predefined uous.
path. For a small fixed-wing UAV, only high-level navigation Assumption 3: Omit some finite time point, system (3) is
control is considered here and low-level attitude control is generalized Lipschitz, that is
achieved via an autopilot system.
|∆y(k + 1)| ≤ L|∆u(k)|, f or∆u(k) ̸= 0 (4)
The UAV dynamics with low-level control in a 2-D hori-
zontal plane is[13] where L is a constant, ∆y(k + 1) = y(k + 1) − y(k) and
∆u(k) = u(k) − u(k − 1) are change of output and input,
ẋ = V cos(ψ)
respectively.
ẏ = V sin(ψ) (1) Remark 1: Assumption 1 is common for many reality
ψ̇ = u systems. Assumption 2 is a typical restrained condition for
where, (x, y) ∈ R2 denote the position of a UAV, V is the general nonlinear system. Assumption 3 can be treated as some
velocity of UAV in the forward direction, ψ is the yaw of limitation to the rate of the change of control energy. These
UAV. The definitions of those variables shown in Fig.1. assumptions are reasonable and acceptable.
The main objective of proposed controller is existing a op- Under assumption 1-3, existing a ϕ(k), called PPD, then
timal control input u(k), which make the following achieved. ∆y(k + 1) = ϕ(k)∆u(k), f or∆u(k) ̸= 0 (5)
|δ(k) − δd | < ε where |ϕ(k)| ≤ L is satisfied for all time.
(2)
umin < u(k) < umax System (5) is an equivalent dynamic linearization presenta-
√ tion of controlled system, and it is distinctly different to model
where, δ(k) = (x(k) − xd (k))2 + (y(k) − yd (k))2 is
linearization method.
distance between the UAV position and desired position,
Consider control cost function [16]
(xd (k), yd (k)) is desired position in time k. δd is desired value
of distance, ε is endurable bound of error, umin and umax are J(u(k)) = |y ∗ (k + 1) − y(k + 1)|2 + λ|u(k) − u(k − 1)|2 (6)
where λ > 0 is a weight factor. y ∗ (k + 1) is desired output. From (7), we can obtain
Substituting (5) into (6), differentiating (6) with respect to ρ1 ϕ1 (k)
u(k) and make it equal to 0. ψ̃d (k) = ψ̃d (k − 1) + (δd (k + 1) − δ(k)) (15)
λ1 + |ϕ1 (k)|2
ρϕ(k)
u(k) = u(k − 1) + (y ∗ (k + 1) − y(k)) (7) where ρ1 ∈ (0, 1] is a step factor.
λ + |ϕ(k)|2 Cost function of estimation of PPD can be selected
where ρ ∈ (0, 1] is a step factor. J(ϕ1 (k)) = |δ(k) − δ(k − 1) − ϕ1 (k)∆ψ̃d (k − 1)|2
PPD ϕ(k) is, obviously, a time-varying parameter even (16)
µ1 |ϕ1 (k) − ϕ̂1 (k − 1)|2
though the controlled system is a time-invariant system. For
estimating PPD, a cost function is also proposed in [16] From (9), it is also obtained
η1 ∆ψ̃d (k−1)
J(ϕ(k)) = |y(k) − y(k − 1) − ϕ(k)∆u(k − 1)|2 ϕ̂1 (k) = ϕ̂1 (k − 1) + µ1 +|∆ψ̃d (k−1)|2
(8) (17)
µ|ϕ(k) − ϕ̂(k − 1)|2 (∆δ(k) − ϕ̂1 (k − 1)∆ψ̃d (k − 1))
solve (8) for maximum or minimum with respect to ϕ(k) the reset condition is same to Remark 2.
η∆u(k−1) For ψ-subsystem, ψ̃(k) and u(k) can be treated as output
ϕ̂(k) = ϕ̂(k − 1) + µ+|∆u(k−1)| 2
(9) and input of system, respectively. The control input u(k) and
(∆y(k) − ϕ̂(k − 1)∆u(k − 1)) estimation of PPD ϕ2 (k) can be obtained using the same
where, λ > 0 is a punishment factor, η ∈ (0, 1] is a step factor. approach in δ-subsystem.
ρ2 ϕ2 (k)
u(k) = u(k − 1) + λ2 +|ϕ2 (k)|2
III. C ONTROLLER D ESIGN ρ1 ϕ1 (k)
(ψ̃d (k − 1) + λ1 +|ϕ1 (k)|2 (δd (k + 1) − δ(k)) − ψ̃(k))
In this section, the data-driven control strategies for fixed-
(18)
wing UAV path following in 2-D and 3-D are obtained. For
ϕ̂2 (k) = ϕ̂2 (k − 1) + µ2η+|∆u(k−1)|
2 ∆u(k−1)
2
system (2), control input is u(k) and δ(k) is output. Assume (19)
desired output is δ ∗ (k), the error of output is defined as (ψ̃d (k) − ϕ̂2 (k − 1)∆u(k − 1))
So that, a date-driven control strategy has been extended to
e1 (k) = δ ∗ (k) − δ(k) (10)
fixed-wing UAV path following problem in 2-D only using the
Lemma 1[16]: For the system (2) using the MFAC algorithm I/O data.
(7) and (9), the expectation of PPD estimated value E(ϕ̂(k)) Remark 3: The control method (19) is directly obtained
is bound. via I/O data of UAV. The process of controller designation
From Lemma 1, proper constants b1 and b2 can be defined is independent to any acknowledge of (13). Furthermore, (19)
as can be updated using each group of data step by step.
0 < b1 < ϕ̂(k) < b2 (11) Remark 4: Although the process of proposed method is
divided into two subsystems, the control input (19) is obtained
The height of fixed-wing UAV is assumed to be constant directly. So there is no problem in time accordance and control
in problem of path following in 2-D plane. Using the discrete rate of different subsystem.
error dynamics are obtained in [17]
IV. S IMULATION
δ(k + 1) = δ(k) + ∆T V sin(ψ̃(k))
(12) In this section, two examples are used to verify our control
ψ̃(k + 1) = ψ̃(k) + ∆T u(k)
algorithm (19) proposed in former section.
where, ∆T is sample time. A nonlinear system is employed for collecting I/O data and
The system (13) can be considered into two subsystem: δ- without additive to control strategy.
subsystem and ψ-subsystem. ẋ = V cos(ψ)
For δ-subsystem, ψ̃(k) and δ(k) can be treated as input and ẏ = V sin(ψ) (20)
output of system, respectively. Refer to (5), this data model
ψ̇ = u
can be presented as
The desired path is from (0,0) to (300,300) in xy plane. The
∆δ(k + 1) = ϕ1 (k)∆ψ̃(k) (13) initial position of the UAV is (0,0) and initial value of heading
where, ∆δ(k+1) = δ(k+1)−δ(k), ∆ψ̃(k) = ψ̃(k)− ψ̃(k−1), angle is 0. The value of parameters in control strategy (15)
ϕ1 (k) is PPD of system. and update law (10) is given η1 = η2 = 0.2, µ1 = µ2 = 1,
Assume existing a proper input ψ̃d (k), which make δ(k) ρ1 = ρ2 = 0.6, λ1 = 0.1, λ2 = 1, ϕ1 (1) = ϕ2 (1) = 1,
asymptotically converge to δd (k). Consider the cost function V = 10, ε = 10− 5, sampling time T = 0.01s.
as following In Fig. 3, results of data-driven control method is shown.
The path of the UAV, represented by a blue solid line,
J(ψ̃d (k)) = |δd (k + 1) − δ(k + 1)|2 + λ1 |ψ̃d (k) − ψ̃d (k − 1)|2 converged to the desired line, donated by a black dot line.
(14) Tracking error and heading angle of the UAV are shown in
 However, since the priori acknowledge of UAV is omitted,
the initial value of this method is cautious to set. Meanwhile,
the results rarely supply guidance to adjustment of parameters.
It is needed to considered those problems in the future work. A
alternative scheme is that combining data-driven control and
model based control. For some loops, data-driven control is
utilized to settle unmodeled parts or uncertain parameters; for
some loops, model based control is utilized to drive the system
efficiently and rapidly.
R EFERENCES
[1] J. Osborne and R. Rysdyk, Waypoint guidance for small UAVs in wind,
Fig. 2. The path of the UAV In Proc. AIAA Infotech Aerospace, Arlington, VA, vol. 193, nos. 1-4, pp.
1-12, 2005.
[2] M. Sun, R. Zhu, and X. Yang, UAV path generation, path following and
gimbal control, in Proc. IEEE Int. Conf. Networking, Sensing Control,
Fig.4 and Fig.5, respectively. The tracking error and heading pp. 870-873, 2008.
angle converge to about 0 and π/4 with some oscillations [3] R. W. Beard and T. W. McLain, Small Unmanned Aircraft: Theory and
due to UAV cross the desired path. Although it is slow to Practice, 1st ed. Princeton, NJ: Princeton Univ. Press, 2012.
[4] Y. Li, N. Sundararajan, P. Saratchandran, Z. Wang. Robust neuro-H∞
converge to the desired path, the tracking error is under 1 controller design for aircraft auto-landing. IEEE Trans. Aerosp. Electron.
from beginning to end. Syst., vol. 40, no. 1, pp. 158-167, 2004.
[5] X. K. Wang, J. H. Qin,C. B. Yu,ISS Method for Coordination Control
of Nonlinear Dynamical Agents Under Directed Topology, IEEE Trans.
Cyber., vol. 44, no. 10, pp. 1832-1845. 2014.
[6] V. Cichella, E. Xargay, V. Dobrokhodov, I. Kaminer, A. M. Pascoal, and
N. Hovakimyan, Geometric 3D path-following control for a fixed-wing
UAV on SO(3), In Proc. AIAA Conf. Guidance, Navigation Control, Aug.
pp. 3578-3592, 2011.
[7] P. B. Sujit, S. SariPalli, and J. B. SouSa, A survey and analysis of
algorithms for fixed-wing unmanned aerial vehicles. IEEE Control System
Magazine, pp. 42-59, 2014.
[8] A. P. Aguiar, J. P. Hespanha, and P. V. Kokotovi, Performance limitations
in reference tracking and path following for nonlinear systems, Automat-
ica, vol. 44, no. 3, pp. 598-610, 2008.
[9] W. J. Rugh and J. S. Shamma, Research on gain-scheduling, Automatica,
vol. 36, no. 10, pp. 1401-1425, 2000.
[10] D. Angeli and E. Mosca, Lyapunov-based switching supervisory control
of nonlinear uncertain systems, IEEE Trans. Autom. Control, vol. 47, no.
Fig. 3. Tracking error of the UAV 3, pp. 500-505, 2002.
[11] D. R. Nelson, D. B. Barber, T. W. McLain, and R. W. Beard, Vector Field
Path Following for Miniature Air Vehicles. IEEE Trans. Robot. Autom.,
vol. 23, no. 3, pp. 519-529, 2007.
[12] A. P. Aguiar, and J. P. Hespanha, Trajectory-Tracking and Path-
Following of Underactuated Autonomous Vehicles With Parametric Mod-
eling Uncertainty. IEEE Trans. Autom. Control, vol. 52, no. 8, pp. 1362-
1379, 2007.
[13] L. R. Carrillo, G. R. Colunga, G. Sanahuja, and R. Lozano, Quad
Rotorcraft Switching Control: An Application for the Task of Path
Following. IEEE Trans. Contr. Syst. Technol., vol. 22, no. 4, pp. 1255-
1267, 2014.
[14] L. Lapierre and B. Jouvencel, Robust Nonlinear Path-Following Control
of an AUV, IEEE J. Ocean. Eng., vol. 33, no. 2, pp. 89-102, 2008.
[15] S. A. Arogeti, and N. Berman, Path Following of Autonomous Vehicles
in the Presence of Sliding Effects, IEEE Trans. Veh. Technol., vol. 61,
no. 4, pp. 1481-1492, 2012.
[16] Z. Hou , X. Bu. Model free adaptive control with data dropouts, Expert
Systems with Applications, pp:10709-10717. 2011.
Fig. 4. Heading angle of the UAV [17] Y. Kang and J. K. Hedrick, Linear Tracking for a Fixed-Wing UAV
Using Nonlinear Model Prective Control, IEEE Trans. Control Syst.
Technol., vol. 17, no. 5, pp:1202-1210, 2009.
[18] R. W. Beard, J. Ferrin, and J. Humpherys, Fixed Wing UAV Path
V. C ONCLUSION Following in Wind With Input Constraints, IEEE Trans. Control Syst.
In this paper, a novel data-driven control is used as a Technol., vol. 22, no. 6, pp. 2103-2117, 2014.
[19] C. Treesatayapun, A Data-driven Adaptive Controller For A Class of Un-
navigation controller of a small fixed-wing UAV. By employ- known Nonlinear Discrete-time System With Estimated PPD, Engineering
ing online I/O data of UAV, the control strategy is proposed Science and Technology, an International Journal, 2015, in press.
directly so that the path of UAV converge to desired path in
the end. Although there are some oscillations in the beginning,
the tracking error is small and converge to about 0.

View publication stats