IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO.

12, DECEMBER 2014 6915

Minimum-Time Trajectory Planning for

Underactuated Overhead Crane Systems
With State and Control Constraints
Xuebo Zhang, Member, IEEE, Yongchun Fang, Senior Member, IEEE, and Ning Sun, Student Member, IEEE

Abstract—In this paper, we propose a novel offline minimum- In this situation, a constrained minimum-time trajectory plan-
time trajectory planning (MTTP) approach for underactuated ning (MTTP) problem naturally arises, that is, how do we find
overhead cranes. To the best of our knowledge, it is the first the minimum-time trajectory to transport the payload from the
optimal solution to the MTTP problem for overhead crane sys-
tems, which simultaneously takes into account various constraints, initial location to the desired one under constraints, including
including the bounded swing angle for the payload, bounded bounded swing angle, bounded velocity, bounded acceleration,
velocity, acceleration, and even jerk for the trolley. Different from and even bounded jerk? The study in this paper aims to provide
existing approaches, by means of system discretization and aug- a satisfactory solution to this practical and important problem.
mentation, the quasi-convex optimization technique is successfully To reduce the oscillatory response during the transportation
adopted to find the minimum-time solution while satisfying all
the aforementioned constraints. Extensive simulation and exper- process, many control techniques have been proposed in the
iments with comparisons to previously published methods are literature, such as input shaping [8], [9], energy-based control
conducted to show the superior performance of the proposed [4], saturation control [11], sliding-mode control [12]–[14],
method. Note that the results derived in this paper also serve as genetic algorithm-based control [15], neural network-based
promising guidance in engineering applications, since it provides control [16], fuzzy control [17], partial feedback linearization-
a performance limit, namely, the possible highest efficiency for
automatic or manual operation of overhead cranes. based control [18], and so on. Among these techniques, input
shaping is essentially an open-loop control strategy, and hence,
Index Terms—Minimum-time trajectory planning (MTTP), it is easy to implement and has been applied to many crane
overhead cranes, underactuated systems.
systems with great success [8], [9]. Sun et al. [4] and Sun and
I. I NTRODUCTION Fang [5] proposed a series of energy coupling feedback control
methods to guarantee the system stability. Several different

U NDERACTUATED mechatronic systems have been

widely applied in modern industries [1]–[3]. As a typical
kind of underactuated systems, overhead crane systems play
sliding-mode control methods are proposed to help accomplish
the antiswing transportation task [12]–[14]. Note that these
works mainly concentrate on how to suppress the undesirable
an important role in many industrial areas to accomplish pay- payload swing, whereas little effort has been made to reveal
load transportation tasks [4]–[7]. For such tasks of overhead time-optimal solutions under various other constraints.
crane systems, a well-known fact is that the transportation In addition to the antiswing requirement, some other practi-
efficiency and the payload antiswing requirement are generally cal physical limits, such as maximum velocity and maximum
contradictory with each other. If the trolley motion is too acceleration specified by the motor manufacturers or users,
fast with large acceleration and velocity, then the undesirable should be also taken into account for a rational control design
payload swing usually appears, even possibly leading to serious of underactuated crane systems. In the literature [22], [23], sev-
accidents such as collisions, and so on. On the other hand, eral time-optimal control methods for crane and other flexible
the pursuit of the highest efficiency under safety requirements systems are presented to improve the transportation efficiency.
and motor performance limits is always a persistent drive for In [22], the overhead crane system is first linearized, and then,
manufacturers to enhance their competitiveness in the markets. bang–bang acceleration control is proposed, wherein the swing
angle is constrained to be zero only at the boundary states.
Manuscript received February 19, 2013; revised October 14, 2013; accepted Nevertheless, there is no guarantee for the swing angle and the
February 28, 2014. Date of publication April 25, 2014; date of current version
September 12, 2014. This work was supported in part by the National Natural velocity to be in a tolerable interval during the transportation
Science Foundation of China under Grant 61203333 and Grant 61325017, in process. In [23], several robust time-optimal control methods
part by the Specialized Research Fund for the Doctoral Program of Higher for flexible systems, which present similar structures with crane
Education of China under Grant 20120031120040, in part by the Tianjin
Natural Science Foundation under Grant 13JCQNJC03200, and in part by the systems, are analyzed; and it is shown that they are equivalent
Open Project of Chongqing Key Laboratory of Computational Intelligence to the nonrobust time-optimal control of some different sys-
under Grant CQ-LCI-2013-03. tems. Indeed, these aforementioned works successfully yield
The authors are with the Institute of Robotics and Automatic Information
System and the Tianjin Key Laboratory of Intelligent Robotics, Nankai Uni- several (robust or nonrobust) minimum-time control techniques
versity, Tianjin 300071, China (e-mail: [email protected]; yfang@ satisfying the boundary conditions; however, they do not take
robot.nankai.edu.cn; [email protected]). into account the safety or physical constraints such as the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. limits of the swing angle and the trolley velocity during the
Digital Object Identifier 10.1109/TIE.2014.2320231 motion process.

0278-0046 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
6916 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 12, DECEMBER 2014

Some minimum-time control methods also incorporate state

constraints in the control design [21], [24]. The work in [24]
converts the problem to a parametric optimization problem to
obtain an optimal solution; however, the controller structure and
the switching time are assumed to be empirically known, and
hence, the consequent optimality is local, i.e., it only holds in
a specified parametric space. Loock et al. utilized the property
of differential flatness and fourth-order B-splines to optimize
the trajectory in the parametric space [21]. Unfortunately, the
velocity bounds are not taken into consideration, and moreover,
the optimality is also local, which only holds in a fourth-order Fig. 1. Overhead crane system model.
B-spline parametric space. Therefore, the MTTP, considering
all state and control constraints, is still a challenging problem. 2) It should be emphasized that the proposed approach is
For controller design of underactuated crane systems, the of significant importance in real industrial applications,
recently proposed motion planning-based control approach [19] since it provides a performance limit, namely, the possi-
first plans the motion trajectory of the trolley, and then, it ble highest efficiency, for automatic or manual operation
utilizes tracking control techniques to accomplish the trans- of overhead cranes. This can be further utilized to evalu-
portation task. Since the motion trajectory is generated by ate other online planning and control strategies or even
known planning techniques, the sketch of the state transition manual operation techniques of trained workers under
process is then also known during the transportation as the safety and physical constraints.
tracking error is generally small under feedback control. The 3) Since no specific control or trajectory structures are as-
merit with generally known trajectories enhances the operation sumed known to convert the original problem into an opti-
safety, and the planning-based control is thus widely used in mization problem over a corresponding parametric space,
various areas such as industrial robots, unmanned vehicles, and the consequent planned trajectory is globally optimal.
so on. However, the existing planning control techniques in The remainder of this paper is organized as follows.
[19] cannot handle all the state constraints such as velocity and Section II presents the continuous-time crane model. In
swing angle bounds. Although the previous work in [20] can Section III, the acceleration-driven optimal planing technique
cope with most constraints for a crane system, it yields only a is presented. In Section IV, we extend the proposed method to
feasible solution rather than a time-optimal one. the jerk-driven situation. Simulation and experimental results
In this paper, an MTTP method is proposed to yield the high- are provided in Section V. Finally, the conclusion is given in
est transportation efficiency under bounded state and control Section VI.
inputs, including the swing angle, trolley velocity, acceleration,
and even jerk. To the best of our knowledge, it is the first
MTTP method for such kind of underacutated systems subject II. S YSTEM D ESCRIPTION
to both state and control constraints. Different from existing Consider a planar overhead crane system, as shown in Fig. 1.
approaches, we use the theory from quasi-convex optimiza- Let x(t) ∈ R and θ(t) ∈ R denote the trolley displacement and
tion to guarantee that the aforementioned constraints can be the payload swing angle with respect to the vertical direction,
satisfied. Specifically, we first discretize the continuous-time respectively, with l being the rope length.
system to obtain a discrete-time acceleration-driven system According to Newton’s second law, it is shown that the
model, based on which the minimum-time criterion is shown to motion of an underactuated crane system is governed by the
be quasi-convex with respect to the optimal control sequence. following equation [20]:
Subsequently, the constraints are analyzed to be convex, and
then, the trajectory planning problem is formulated into a lθ̈ + cos θẍ + g sin θ = 0 (1)
constrained quasi-convex optimization problem. Using the bi-
section method, the solution converges to the optimal one by where the constant g ∈ R is the gravity acceleration.
solving a convex feasibility problem at every step of search. Since the proposed method can guarantee that the swing
In addition, by augmenting the system state, the jerk-driven angle is always in a small range by using the quasi-convex op-
system model is derived, and then, the proposed method is timization techniques during the whole transportation process,
extended to further cope with the jerk constraint, as well as the we use the standard small angle approximations of sin θ ≈ θ
constraints on the swing angle, the velocity, and acceleration. and cos θ ≈ 1 as usually adopted in the literature [8], [20]–[22],
Extensive simulation and experimental results with compar- [24]. In this case, the system equation is rewritten as
isons to previously published methods are provided to show the g 1
superior performance of the proposed method. θ̈ + θ = − ẍ. (2)
l l
Compared with existing approaches, the main contribution
lies in the following. From this equation, it is clear to see that the motion of the
1) The optimal minimum-time solution successfully han- trolley (the acceleration ẍ) directly influences the evolution
dles bounded state and control constraints, including the of the payload swing angle θ. Therefore, it is important to
swing angle, trolley velocity, acceleration, and even jerk. elegantly design the trolley motion to obtain a highly efficient
ZHANG et al.: MTTP FOR OVERHEAD CRANE SYSTEMS WITH STATE AND CONTROL CONSTRAINTS 6917

III. MTTP W ITH C ONTINUOUS V ELOCITY

To achieve the global optimality, one way is to formulate
the problem as a convex or quasi-convex optimization problem.
By discretization of a continuous-time system, we first derive a
discrete-time dynamic system, based on which it is shown that
the minimum-time optimal criterion is quasi-convex, and other
state/control constraints are convex with respect to the control
Fig. 2. Overall block diagram for planning and control of overhead crane sequences to be designed. Therefore, the MTTP problem is
systems.
successfully converted to a quasi-convex optimization problem.
Then, the bisection method is utilized to find the optimal
trajectory under various constraints such as bounded swing an-
solution. It should be noted that, here, the control input of the
gle, bounded velocity, bounded acceleration, and even bounded
system is the acceleration; hence, the velocity is guaranteed
jerk, and so on.
to be continuous, since it is the integral of the acceleration
In industrial applications, a typical crane task is to transport
profile.
the payload horizontally from an initial position x(t0 ) = x0 to
a target position x(tf ) = xf under safety and physical con-
straints, with t0 , tf ∈ R being the initial time and final arrival A. Preliminaries
time, respectively. Without loss of generality, suppose t0 = 0 in
the following analysis. Definition 1—Quasi-convex Function [26]: A scalar func-
By considering requirements for real transportation tasks, tion f (ρ) : Rn → R is called quasi-convex (or unimodal) with
trajectory planning of overhead crane systems here is defined respect to ρ ∈ Rn if its domain and all its sublevel sets Sα
as to find feasible acceleration (or jerk) trajectories such that
Sα = {ρ ∈ domf | f (ρ) ≤ α} (3)
the resultant trolley position trajectory smoothly evolutes from
x(0) = x0 to x(tf ) = xf , and that the resultant swing angle,
for α ∈ R, are convex. The notation domf denotes the domain
velocity, acceleration, and even jerk are kept in specified (ex-
for the function f .
plicitly bounded) ranges during the motion process.
Definition 2—Quasi-convex Optimization Problem [26]: A
Accordingly, the MTTP aims to find among all the feasible
quasi-convex optimization problem is one of the form
trajectories, the optimal one, which minimizes the final arriving
time tf . Note that the minimum-time optimal trajectory plan- minimize f0 (ρ) (4)
ning will definitely improve the transportation efficiency for
an automatic overhead crane system, and on the other hand, subject to fi (ρ) < 0, i = 1, 2, . . . , m (5)
it also provides a practical performance limit of the manual
operation, which can be used to evaluate the efficiency of aTi ρ = bi , i = 1, 2, . . . , p (6)
human operators.
To further clarify the function of the minimum-time trajec- where the objective function f0 is a quasi-convex function, and
tory planner, the overall block diagram for planning and control f1 , f2 , . . . , fm are convex functions for a number of m convex
of overhead crane systems is provided in Fig. 2. The whole constraints, with ai and bi being constant known vectors for a
system is composed of two modules, i.e., a trajectory plan- number of p linear constraints.
ning module and a tracking control module. The work of this In addition to the aforementioned concepts, we should also
paper mainly focuses on the minimum-time offline trajectory keep in mind that any quasi-convex optimization problem can
planning module. The inputs of this module include initial and be converted into solving a sequence of convex feasibility prob-
desired configurations, an acceleration-based kinematics model lems, and thus, the globally optimal solution can be obtained by
or a jerk-based kinematics model(which will be introduced using the bisection method, as stated in [26]. Therefore, once
by system augmentation in Section IV), and various state and the problem is successfully formulated as a standard convex
control constraints. The outputs of the planner consist of all the optimization algorithm, it can be solved using the bisection
state and control trajectories, including the position trajectory, method, wherein the users should carefully shrink the range
the velocity trajectory, the swing angle trajectory, the accel- between the lower and upper bounds of the objective function
eration trajectory (and the jerk trajectory if using jerk-based to increase the computing efficiency.
kinematics model). However, it should be remarked that, in
general cases, it is unnecessary to choose all of these trajectories
as the reference trajectories for the subsequent controller to B. System Discretization
track, although they can be arbitrarily selected by interested Here, we first express the overhead crane system in the
users. In this paper, we just choose the position and velocity continuous-time state-space form, based on which the dis-
trajectories as reference trajectories, and the previously devel- cretization can be easily carried out.
oped trajectory tracking controller in literature [10], [20] is To facilitate the subsequent analysis, we rewrite (2) into the
directly adopted in the tracking control module. Consequently, state-space representation as
the torque control inputs are computed in real time and then
issued to track the reference trajectories. q̇ = Aq + Bu (7)
6918 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 12, DECEMBER 2014

with requirement of general transportation tasks for overhead crane

systems, it is generally assumed q 0 = [x0 0 0 0]T [19], [20].
q = [x v θ ξ ]T , (8) Namely, the initial trolley velocity v0 , the initial swing angle
⎡ ⎤ ⎤
⎡
0 1 0 0 0 θ0 , and the initial angular velocity ξ0 are all zero.
⎢0 0 0 0⎥ ⎢ 1 ⎥ 1) Analysis of the Minimum-Time Criterion Function: In
A=⎣ ⎦, B=⎣ ⎦ (9)
0 0 0 1 0 this part, we will first show that the minimum-time criterion is
0 0 − gl 0 −l
1
actually a quasi-convex function with respect to the acceleration
where v(t) ∈ R and ξ(t) ∈ R denote the trolley velocity and the input u(k). This is an important result in this paper, which
payload swing angular velocity, respectively; and the control makes it possible to formulate the trajectory planning task as
input u(t) ∈ R represents the acceleration of the trolley. q(t) ∈ a quasi-convex optimization problem.
R4 is referred to as the system state vector, with matrices A ∈ Since the system is controllable, there must exist some
R4×4 and B ∈ R4×1 being the state matrix and the input matrix, feasible control sequences, which can drive the system state
respectively. from q 0 to q f . Let K ∈ R be a sufficiently large integer such
Let T ∈ R denote the sampling period, then after some cal- that the goal configuration can be reached at time KT , then,
culations, we can obtain the corresponding exact discrete-time for every feasible control sequence (u(0), u(1), . . . , u(K)), we
model of the computer-based control system under the zero- define the planning time ft (·) as follows:
order hold assumption. Accordingly, the control input u(k) ft (u(0), u(1), . . . , u(K))
remains constant between kT and (k + 1)T , and hence, system = min{kp |q(k) = q f for kp ≤ k ≤ K} (13)
(7) is discretized as follows:
which indicates that the overhead crane keeps still after the
q(k + 1) = G(T )q(k) + H(T )u(k) (10) planning time. To show that function ft (·) is quasi-convex with
respect to variables (u(0), u(1), . . . , u(K)), given an arbitrary
with
constant fs ∈ Z+ , the sublevel set Sa derived by ft (·) ≤ fs is
T defined as follows:
G(T ) = eAT , H(T ) = eAt dtB (11)
Sa = (u(0), . . . , u(K)) |q(k) = q f for any fs ≤ k ≤ K . (14)
0

where G(T ) ∈ R4×4 and H(T ) ∈ R4×1 are the corresponding Since the trolley remains stationary at the desired position, it
discrete-time state matrix and input matrix for a fixed sampling can be seen that once the final configuration is arrived, the
time T , respectively. It follows from (10) that acceleration input should be kept as zero from then. By further
using the state transition (12), we have
q(k) = Gk q 0 + Gk−1 Hu(0) + · · · + Hu(k − 1)
k
Sa = (u(0), . . . , u(K)) |u(k) = 0 for fs ≤ k ≤ K
= Gk q 0 + Gk−i Hu(i − 1) (12)
i=1 fs
Gfs q 0 + Gfs −i Hu(i − 1) = q f . (15)
which shows the state transition process, and it is clear that
i=1
the system state at a specified time kT is affine to the ex-
erted control sequence (u(0), u(1), . . . , u(k − 1)). Hence, we Based on the fact that the solution set of linear equations is
conclude that, by means of system discretization, the linear always affine, we know from (15) that the sublevel set Sa is
constraints on the system state can be further transformed into an affine set with respect to the control sequence. Since every
linear constraints with respect to the input control sequences affine set is also convex, it is concluded that the sublevel set
(u(0), u(1), . . . , u(k − 1)). Sa is convex, which indicates that the function ft (·) is quasi-
convex with respect to the control sequence to be optimized
C. Quasi-convex Optimization-Based (u(0), u(1), . . . , u(K)) based on the definition of the quasi-
Minimum-Time Planning convex function in Section III-A.
With the discrete-time system model, the aim of the MTTP 2) Analysis of State and Control Constraints: To ensure
now becomes to find the optimal control input sequences, safety transportation, the swing angle θ(k) should satisfy the
namely, the acceleration profile u(k), to steer the system state following constraint C1:
from an initial state q 0 = [x0 v0 θ0 ξ0 ]T to the desired final state
C1 : |θ(k)| ≤ θmax (16)
q f = [xf 0 0 0]T under state and control constraints, including
bounded swing angle, velocity, and acceleration. According where θmax is the maximum allowable swing angle. In addition,
to the derived trajectory for the control input, those state since the motor has its own performance limits provided by
trajectories, including the position, velocity, and swing angle the manufacturers, velocity v(k), and acceleration u(k) should
trajectories, can be then obtained using the state transition (10). meet the following requirements:
Note that the proposed approach is able to handle the planning
problem from any initial state to the desired state with θ0 C2 : |v(k)| ≤ vmax (17)
being a small swing angle. Nevertheless, consider the practical C3 : |u(k)| ≤ umax (18)
ZHANG et al.: MTTP FOR OVERHEAD CRANE SYSTEMS WITH STATE AND CONTROL CONSTRAINTS 6919

where vmax and umax denote the maximum velocity and accel- So far, the minimum-time constrained trajectory planning prob-
eration. {Using the result obtained by the system discretization lem is converted into a constrained optimization problem for-
in (12), we know that these constraints can be further trans- mulated by (31), (32) and (23)–(27). In the following, we will
formed into linear constraints with respect to the input control show that this problem is essentially a quasi-convex optimiza-
sequences (u(0), u(1), . . . , u(k − 1)). Specifically, since the tion problem. By the definition of the quasi-convex function in
position, velocity, and the swing angle are all entries of the Section III-A, we have shown that the objective function ft (·)
system state q(k), we have in (31) is quasi-convex with respect to the control sequence
(u(0), u(1), . . . , u(K)). In addition, the constraints for the
v(k) = I Tv q(k), θ(k) = I Tθ q(k) (19) swing angle θ(k), velocity v(k), and the control input u(k)
are all linear constraints of the control sequence u(k)(k =
with I v and I θ being unit vectors as follows:
0, 1, . . . , K), as shown in (23)–(27). Therefore, the constraints
I v = [0, 1, 0, 0]T , I θ = [0, 0, 1, 0]T . (20) (32) and (23)–(27) are all standard convex constraints. To sum
up, the objective function is quasi-convex and other constraints
By further substituting (12) into (19), we know that are convex; hence, it is concluded that the problem formu-
k lated by (31), (32) and (23)–(27), is essentially a quasi-convex
v(k) = I Tv Gk q 0 + I Tv Gk−i Hu(i − 1) (21) optimization problem according to definition 2 provided in
i=1 Section III-A.
k 4) Solving the Quasi-convex Optimization Problem Using
θ(k) = I Tθ Gk q 0 + I Tθ Gk−i Hu(i − 1). (22) Bisection: The solution of a quasi-convex optimization prob-
i=1 lem can be obtained by using the bisection method, together
Hence, the aforementioned state and control constraints in with solving a sequence of convex feasibility problems. To
(16)–(18) can be converted into linear constraints with respect describe the approach in a self-contained way, we provide the
to (u(0), u(1), . . . , u(k − 1)) as follows: detailed computation process in Algorithm 1. In this algo-
rithm, k  ∈ Z+ denote the minimum time index, with top =
k
k  T being the minimum time for the presented constrained
I Tθ Gk q 0 + I Tθ Gk−i Hu(i − 1) ≤ θmax (23) trajectory planning. Given the lower bound kl ∈ Z+ and the
i=1
k
upper bound kup ∈ Z+ for k  , the minimum time index k 
−I Tθ Gk q 0 − I Tθ Gk−i Hu(i − 1) ≤ θmax (24) and the corresponding optimal control sequence uop (k)(k =
i=1
1, 2, . . . , k  ) with uop (k) = 0(k  < k < K) can be obtained
k using the bisection method.
I Tv Gk q 0 + I Tv Gk−i Hu(i − 1) ≤ vmax (25)
i=1
k Algorithm 1 Bisection-based quasi-convex optimization for
−I Tv Gk q 0 − I Tv Gk−i
Hu(i − 1) ≤ vmax (26) MTTP of overhead cranes
i=1
u(k) ≤ umax , −u(k) ≤ umax . (27) Input: q 0 , q f , kl , kup , G(T ), H(T ), vmax , θmax , umax
Output: k  , uop (k)(k = 0, 1, . . . , k  )
3) Quasi-convex Optimization Formulation: Now, we can 1: repeat
formulate the MTTP problem as finding the optimal control se- 2: set kmid = (kl + kup )/2
quence uop (k)(k = 0, 1, 2, . . . , K) among all feasible sequences 3: solve the convex feasibility problem
to minimize the planning time ft (u(0), u(1), . . . , u(K))
Find u(k)(k = 0, 1, . . . , kmid ) (33)
kmid
minimize, ft (u(0), u(1), . . . , u(K)) (28)
s.t. Gkmid q 0 + Gkmid −i Hu(i − 1) = q f ;
s.t. q(0) = [x0 v0 θ0 ξ0 ]T ;
i=1
For ft ≤ k ≤ K, q(k) = q f (29) For 0 ≤ k ≤ kmid , linear constraints (23)–(27). (34)
For 0 ≤ k ≤ K,
θ(k) ≤ θmax , v(k) ≤ vmax , u(k) ≤ umax . (30) 4: if the problem is feasible then
5: kup = kmid
By using the fact that the trolley will stop and keep still at the
6: else {the problem is infeasible}
desired position, after some mathematical analysis, the previous
7: kl = kmid
optimization problem (28)–(30) is reformulated as
8: end if
minimize ft (u(0), u(1), . . . , u(K)) (31) 9: until kl − kup ≥ 0
s.t. q(0) = [x0 v0 θ0 ξ0 ]T ; 10: k  = kl , and uop (k) = u(k) for k = 0, 1, . . . , k  .
For ft ≤ k ≤ K, u(k) = 0;
ft
Gft q 0 + Gft −i Hu(i − 1) = q f ; The “Find” operation in Algorithm 1, together with the
i=1 state and control constraints by (34) and (23)–(27), constitutes
For 0 ≤ k ≤ K, a feasibility problem for convex optimization. Note that the
linear constraints for u(k), v(k), θ(k) in (23)–(27). (32) objective function is empty, and the objective of the undergoing
6920 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 12, DECEMBER 2014

computation is reduced to find a feasible solution for variables

u(k)(k = 1, 2, . . . , kmid ) subject to the constraints from (34)
and (23)–(27). In fact, this convex feasibility problem is essen-
tially a linear programming (LP) problem with an empty objec-
tive function, because the constraints from (34) and (23)–(27)
are all linear with respect to the decision variables u(k)(k =
1, 2, . . . , kmid ). Many LPs or more general convex program-
ming methods, such as interior-point methods, can be utilized
to solve this feasibility problem. These methods are included in
many available free or commercial optimization software.
To increase the search efficiency, it is generally a good way to
shrink the search range by providing a lower bound and upper
bound of the arrival time k  T that are close to the optimal
solution. In the algorithm, an initial estimation of the lower
bound for the time index kl is given as
xf
kl = (35)
vmax T
whereas the upper bound kup is regarded as the arrival time
determined by the online trajectory generating method [10].
It is worthwhile to point out that once the optimal trajectory
for the control input is obtained, we can use the state transition
(12) to compute all the state trajectories, including the position,
velocity, and swing angle trajectories. Any of these trajectories
can be utilized as reference trajectories to be followed by a
feedback tracking controller, as shown in Fig. 2.
Fig. 3. Comparison of the proposed method and existing approaches without
IV. E XTENSION TO M INIMUM -T IME P LANNING jerk constraint. (Blue solid lines) Results using the proposed MTTP method.
W ITH C ONTINUOUS ACCELERATION (Red dashed lines) Results by BBOC [22]. (Green dotted-dashed lines) Results
by PPBM1 [20].
Here, we will show that the proposed methodology can be
extended to ensure that the trolley acceleration is continuous in
discrete-time sense (i.e., the jerk is bounded), which is bene- time system model and then formulating the minimum planning
ficial to the actuator to avoid undesired vibration. To address problem as a quasi-convex optimization problem. In addition,
this problem, we will first derive an augmentation system by the bounded jerk constraint can be also incorporated
adding the acceleration as a state variable. In this case, the
crane system is jerk driven, and thus, we can consider all C4 : |u (k)| ≤ umax . (39)
state and control constraints, including bounded jerk, bounded
acceleration, bounded velocity, and bounded swing angle in a By using a formulation similar to (31), (32), and (23)–(27)
unified framework. with extra jerk constraint (39), the bisection-based optimization
Specifically, define the augmented system state q  ∈ R5 as algorithm can be further applied to obtain the minimum time
index k ∗ and the corresponding control sequence u (k)(k =
q  = [q, a]T (36) 0, 1, . . . , K), as well as the generated trajectory.
Remark 1: By augmenting the system state, the proposed
where a(t) ∈ R denotes the trolley acceleration (a(t) is equiv- approach has the potential capacity to generate minimum-time
alent to u(t) in (7)). In addition, we define the control input trajectory satisfying high order smoothness of the resulting
u (t) ∈ R as the jerk of the trolley motion. Subsequently, the practical control input such as motor speed/torque.
jerk-driven system model is derived as Remark 2: Some transient performance index, such as the
q̇  = A q  + B  u (37) requirement of no overshoot for x(t) (namely, x(k) ≤ xf for
any k), can be also considered in the quasi-convex optimization-
with based framework since it is a convex constraint with respect to
⎡ ⎤ ⎡ ⎤ the control sequence.
0 1 0 0 0 0
⎢0 0 0 0 1 ⎥ ⎢0⎥
⎢ ⎥ ⎢ ⎥
A = ⎢ 0 0 0 1 0 ⎥, B = ⎢ 0 ⎥ . (38)
⎣ ⎦ ⎣ ⎦ V. S IMULATION AND E XPERIMENTAL R ESULTS
0 0 −g/l 0 −1/l 0
0 0 0 0 0 1 A. Simulation
According to the continuous-time jerk-driven model (37), the Here, we present three groups of numerical simulation re-
proposed approach can be applied by first obtaining its discrete- sults for the proposed MTTP in comparison with existing
ZHANG et al.: MTTP FOR OVERHEAD CRANE SYSTEMS WITH STATE AND CONTROL CONSTRAINTS 6921

TABLE I
T RAJECTORY P ROPERTIES

approaches, including bang–bang optimal control (BBOC)

[22], the phase plane-based trajectory planning method (PPBM)
[20] and the newly developed coupling analysis-based trajec-
tory generating method (CABM) [10]. To facilitate subsequent
analysis, the phase plane-based planning method (PPBM) that
does not take into account the bounded jerk constraint is termed
as “PPBM1”, whereas the PPBM that guarantees the bounded
jerk constraint is abbreviated as “PPBM2”.
In the simulation, the rope length and the gravity constant
are set as l = 1.2 m, g = 9.8 m/s2 . The initial and desired con-
figuration are selected as q 0 = [0 0 0 0], q f = [10(m) 0 0 0],
whereas the state and control constraints are as follows:

θmax = 0.1 rad, vmax = 2 m/s, umax = 1 m/s2 . (40)

It should be noted that the maximum swing angle is chosen to

satisfy the requirement of most general applications, whereas
other parameters are set close to those of real industrial over-
head cranes.
In the first simulation, we suppose that the control input is
the acceleration u(t) without the bounded jerk constraint, and
hence the trolley velocity is naturally continuous. Fig. 3 depicts
the simulation results in a comparative way by utilizing various
methods, including BBOC, PPBM1 and the proposed MTTP
method. It is shown that, although the bang–bang time optimal
control presents the fastest convergent speed to the desired
configuration, the state constraints (C1 and C2), including the
prescribed maximum swing angle and the maximum velocity,
are violated during the planning process. Between the rest
two methods that obey the state and control constraints, it is
clear that the proposed method is much more efficient than the
phase plane-based method without the bounded jerk constraint
(PPBM1) [20]. This fact is further illustrated by the arrival time
listed in Table I, from which we can see that the resulting ef- Fig. 4. Comparison of the proposed method and existing approaches with the
ficiency is improved by (11.48 − 7.68)/11.48 = 33.1% when jerk constraint. (Blue solid lines) Results using the proposed MTTP method.
(Red dashed lines) Results by the CABM [10]. (Green dotted-dashed lines)
compared with PPBM1. Hence, the proposed method presents Results by PPBM2 [20].
superior performance over existing approaches in the presence
of the state and control constraints, and the resulting arrival time
can be used as the reference performance limit for automatic or driven BBOC leads to infinite jerks at the switching point, the
manual transportation using crane systems. corresponding simulation results are not included in this group
In the second simulation, we suppose that the jerk should be of simulation. In the following, we will show comparative
bounded during the planning process. Hence, the acceleration is simulation results using the proposed MMTP, PPBM2 [20], and
continuous, and it can be easily implemented by the actuators CABM [10]. As shown in Fig. 4, the constraints (C1-C3) on the
without abrupt switchings of the acceleration in the bang–bang velocity, the acceleration, and the swing angle can be all satis-
control. The maximum jerk constraint C4 in (39) is set as fied using different methods. However, the maximum allowable
jerk constraint (C4) is violated by PPBM2 [20]. The reason is
umax = 2 m/s3 . (41) that, although the resulting jerk by PPBM2 can be proven to be
bounded in [20], the specified upper bound cannot be arbitrarily
To ensure that C4 constraint is satisfied, we use the jerk-driven enforced as an explicit constraint. In addition, it is obvious
system model described in Section IV. Since the acceleration- that the proposed method uses less time, and it is thus more
6922 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 12, DECEMBER 2014

TABLE II
T RAJECTORY P ROPERTIES

Fig. 6. Test bed.

tion under differential constraints of the kinematic equation.

Since the initial velocity is greater than zero (1 m/s), the
resultant trajectories become much faster than those in the
first simulation. This simulation demonstrates that the proposed
approach can potentially act as an online trajectory planner, for
which the initial state can be any nonstatic system state. With
the fast development of computing hardware and optimization
software, it is promising that real-time online implementation
of the proposed planner can be realized in the future.

Fig. 5. Results of MTTP with nonzero initial velocity and swing angle. B. Experiments

efficient than the other two approaches. An underlying reason Here, comparative experimental results are conducted to
is that the resulting acceleration and velocity are attractive since show the superior performance of the proposed approach. The
the capacity of the actuators is employed as much as possible test bed used in the experiments is shown in Fig. 6, where
under safety and physical requirements. The corresponding the rope length is l = 0.75 m, whereas the initial and desired
arrival time using these three methods are also presented in locations are set as x0 = 0 m and xf = 0.6 m. During the
Table II, which further confirms the effectiveness of the pro- planning process, the maximum swing angle, velocity, and
posed method. It is shown that the efficiency of the proposed acceleration are selected as
method is improved by (12.58 − 8.20)/12.58 = 34.82% with
respect to CABM [10], and (11.59 − 8.20)/11.59 = 29.25% θmax = 3.0 deg, vmax = 0.1 m/s, amax = 0.2 m/s2 (42)
with respect to PPBM2 [20].
To further verify that the proposed method works well with whereas the maximum allowable jerk is set as
any initial state, we conduct the third simulation with the initial
velocity and the swing angle being nonzero. In this case, to the umax = 2 m/s3 . (43)
best of our knowledge, most existing methods cannot be di-
rectly applied; thus, we only provide the results of the proposed
approach. Without loss of generality, we use the acceleration Due to the small motion range of our experimental test bed,
u(t) as control input, and the initial and desired configurations the values for the maximum acceleration, velocity, and the
are selected as swing angle are set smaller than those in the simulation such
that these constraints can really have functions during the plan-
q 0 = [0 1 (m/s) 0.1 (rad) 0], q f = [10 (m) 0 0 0] ning process. By taking into account the aforementioned state
and control constraints, various trajectories can be obtained by
with other constraints being the same as those in (40). using different planning methods. Here, two groups of com-
Fig. 5 depicts the trajectory results for position, velocity, parative experimental results are presented, wherein the jerk
acceleration, and the swing angle. It can be shown that the constraint is not considered during the planning process in the
system state is successfully planned to the target configura- first group, whereas it is considered in the second group. Note
ZHANG et al.: MTTP FOR OVERHEAD CRANE SYSTEMS WITH STATE AND CONTROL CONSTRAINTS 6923

TABLE III
T RAJECTORY P ROPERTIES

Fig. 7. Experimental comparison of the proposed MTTP method, PPBM1

[20], and BBOC. (Red dotted lines) Corresponding reference trajectories
computed by MTTP, PPBM1, and BBOC. (Blue solid lines) Trajectory tracking
experimental results based on the proposed MTTP. (Green dotted-dashed lines)
Experimental results by PPBM1. (Black dashed lines) Results by BBOC.

that these trajectory planning methods are employed to gener- Fig. 8. Experimental comparison of the proposed method and existing ap-
proaches with the bounded jerk constraint. (Red dotted lines) Corresponding
ate various reference trajectories, and in this experiment, the reference trajectories computed by MTTP, PPBM2, and CABM. (Blue solid
planned position trajectory and the planned velocity trajectory lines) Results using the proposed MTTP. (Green dotted-dashed lines) Results
are utilized as the reference signal for the subsequent con- by PPBM2. (Black dashed lines) Results by CABM.
troller to track. As previously shown in Fig. 2, a proportional-
derivative (PD) feedback torque controller is applied to track the theoretic results that the BBOC method cannot guarantee
the generated reference trajectories [10], [20] to achieve highly the velocity and swing angle constraints to be satisfied. In addi-
efficient transportation tasks. tion, it is seen that, in the presence of the same state and control
1) Group 1—Experimental Results Without the Jerk Con- constraints, the proposed MTTP method is more efficient than
straint: In this group, the bounded jerk constraint is not en- PPBM2. This fact is also illustrated in Table III, wherein the
forced during the planned process. For comparative analysis, ideal computed arrival time of the reference trajectories and the
the proposed MTTP method based on the acceleration-driven real arrival time in the experiments are listed for comparisons.
model in Section III and PPBM1 [20], are adopted to generate It is seen that the efficiency of the proposed MTTP method is
the reference trajectories under the constraints in (42), and then, improved by (7.910 − 6.935)/7.910 = 12.33% with respect to
the corresponding PD tracking controllers are implemented the PPBM1. It should be noted that some residual swing or
on the test bed to track the trajectories. chattering of the velocity may arise since the jerk is very large
Fig. 7 depicts the experimental results by using MTTP, at the bang–bang-like switching points, which usually leads to
PPBM2, and BBOC. It is shown that the reference position and physical vibration of motor with respect to its base.
velocity trajectories generated by using the MMTP and PPBM2 2) Group 2—Experimental Results With the Jerk Constraint:
approaches are both tracked very well using PD controllers, and To ensure that the jerk lies in a specified interval, we adopt
the swing angle is less than the prescribed maximum allowable the proposed MTTP method based on the jerk-driven system
value of 3◦ . {However, it should be noted that, although the ar- model presented in Section IV. For comparative analysis, other
rival time of the BBOC approach is the shortest, the maximum two approaches, including CABM in [10] and phase plane-
velocity is greater than 0.35 m/s for both reference and practical based method with the jerk constraint (PPBM2) in [20], are also
velocity trajectories, which is obviously more than the specified implemented in real experiments.
maximum velocity 0.1 m/s in (42). The violation of the maxi- Fig. 8 depicts the comparative experimental results, from
mum velocity constraints in the experiment is consistent with which it is shown that the proposed MTTP method is the
6924 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 12, DECEMBER 2014

TABLE IV trajectory still works. In this case, the motor and its
T RAJECTORY P ROPERTIES
drive should be carefully selected and the feedback tra-
jectory tracking controllers should be elegantly designed
to gain more robustness in the presence of payload weight
variation [19].
3) Disturbances: the proposed approach is mainly focused
on trajectory planning, which is actually an open-loop op-
timal planner. The output of the planner will be regarded
most efficient approach under the same state and control con-
as the reference trajectory to be tracked by using a feed-
straints. From Table IV, we know that the efficiency of the pro-
back tracking controller. Although the open-loop planner
posed MTTP method is improved by (7.845 − 6.945)/7.845 =
cannot deal with disturbances, disturbance rejection can
11.47% with respect to PPBM2, and (8.700 − 6.945)/8.700 =
be considered in the tracking controller.
20.17% with respect to CABM.
4) Offline versus online planning: the algorithm is offline
Additionally, it is worthwhile to point out that the proposed
in its current form. Since the algorithm works well with
MTTP method can guarantee the jerk to be less than the spec-
respect to any initial state, the planner can potentially
ified maximum value (43); however, the other two approaches,
act as an online planner, provided that the computation
including PPBM2 and CABM, only ensure that the jerk is
process can be implemented in a real-time manner. In
bounded, yet they cannot guarantee that the jerk is less than
fact, the computing cost is increasing when the sampling
the specified explicit maximum value. This fact is indicated
time is smaller, because in this case, the discretized
by the chattering of the velocity for the PPBM2 and CABM,
control input sequences becomes a large set of vari-
as shown in Fig. 8. For the proposed MTTP method, since
ables. For typical crane applications, by using CVX [27]
the jerk is explicitly limited to a range by using a jerk-driven
and GUROBI [28] in MATLAB, the computing time
system model, it is seen that the vibration of the motor is less
is generally less than 3 min given a sampling period
than that for an acceleration-driven model in the first group of
of 0.02 s, which is acceptable for offline planning but
experiments, and hence, the residual swing is almost negligible.
cannot be directly implemented for online planning. For
Remark 3: Note that the acceleration and jerk profiles are
industrial applications, other than directly implementing
not given in the experiments due to the lack of sensors for
the proposed approach, a possible way is to use the
these variables on our test bed. However, it can be seen that
lookup tables to memorize a set of useful trajectories,
the position and velocity are tracked very well, which further
or to use neural network-based methods to learn from a
indicates that the real acceleration and jerk are generally close
huge set of offline trajectories generated by the proposed
to the reference ones when the noise and disturbance are small.
approach. When the time-consuming training is over, we
Remark 4: Due to the limited motion distance of our test
can then implement these tables or neural networks on
bed, the efficiency improvements by using the proposed MTTP
an industrial portable device in real time. Compared with
in the experiments are less than those in the simulation, wherein
direct implementation, this solution is practical and can
the target distance is set as 10 m such as the real industrial
potentially make the current planner an online planner,
cranes. Hence, the proposed MTTP is of vital importance
under limited computing ability for portable devices.
for efficiency improvement if it is implemented in industrial
applications with long motion distances.
VI. C ONCLUSION

C. Some Notes for Users Most of current researches on motion planning of overhead
cranes aim to obtain feasible trajectories with the bounded
To successfully apply the proposed trajectory planner in swing angle constraint. Few results have been yet presented for
practice, many factors should be taken into account for a minimum-time optimal planning of overhead cranes under con-
realistic transportation task. Parameter variation, including the straints on the swing angle, velocity, acceleration, and even jerk.
changing of rope length and payload weight, usually happens. This paper has proposed an offline MTTP approach for over-
In addition, both external and internal disturbances, such as the head crane systems. An advantage of the proposed approach
wind and highly nonlinear friction, also appear in the motion is that both state and control constraints are successfully taken
process. The possibility of implementing the algorithm in an into consideration during the planning process. Different from
online manner should be also discussed. In order to deal with existing approaches, the presented method adopts quasi-convex
these problems, we provide some notes and suggestions to serve optimization theory to guarantee state and control constraints
as guidance for potential users. to be satisfied. Consequently, the proposed approach gives a
1) Rope length variation: when the rope length changes, the performance limit of the transportation efficiency under safety
trajectory needs to be replanned by running the algorithm requirements, which is of significant importance in engineer-
again. An existing issue is that the current proposed ing applications to evaluate the performance of automatic or
approach is offline. To cope with this issue, please refer to manual operation of cranes. Both simulation and experimental
the subsequent notes on “offline versus online planning.” results with comparisons to existing mainstream approaches are
2) Payload weight variation: since the payload weight does presented to show the superior performance of the proposed
not appear in the system kinematics (1), the planned approach.
ZHANG et al.: MTTP FOR OVERHEAD CRANE SYSTEMS WITH STATE AND CONTROL CONSTRAINTS 6925

R EFERENCES [21] W. Van Loock, G. Pipeleers, and J. Swevers, “Optimal input design for flat
systems using B-Splines,” in Proc. Amer. Control Conf., San Francisco,
[1] Y. Fang, P. Wang, N. Sun, and Y. Zhang, “Dynamics analysis and non-
CA, USA, 2011, pp. 3281–3282.
linear control of an offshore boom crane,” IEEE Trans. Ind. Electron.,
[22] G. A. Manson, “Time-optimal control of an overhead crane model,” Op-
vol. 61, no. 1, pp. 414–427, Jan. 2014.
tim. Control Appl. Methods, vol. 3, no. 2, pp. 115–120, Apr.–Jun. 1982.
[2] Y. Fang, X. Liu, and X. Zhang, “Adaptive active visual servoing of non-
[23] L. Y. Pao and W. Singhose, “Robust minimum time control of flexible
holonomic mobile robots,” IEEE Trans. Ind. Electron., vol. 59, no. 1,
structures,” Automatica, vol. 34, no. 2, pp. 229–236, Feb. 1998.
pp. 486–497, Jan. 2012. [24] A. Z. AI-Garni, K. A. F. Moustafa, and S. Nizami, “Optimal control
[3] M. T. Ravichandran and A. D. Mahindrakar, “Robust stabilization of
of overhead cranes,” Control Eng. Pract., vol. 3, no. 9, pp. 127–1284,
a class of underactuated mechanical systems using time-scaling and
Sep. 1995.
Lyapunov redesign,” IEEE Trans. Ind. Electron., vol. 58, no. 9, pp. 4299–
[25] K. Peng, W. Singhose, and D. Frakes, “Hand-motion crane control using
4313, Sep. 2011. radio-frequency real-time location systems,” IEEE/ASME Trans. Mecha-
[4] N. Sun, Y. Fang, and X. Zhang, “Energy coupling output feedback con-
tronics, vol. 17, no. 3, pp. 464–471, Jun. 2012.
trol of 4-D OF underactuated cranes with saturated inputs,” Automatica,
[26] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
vol. 49, no. 5, pp. 1318–1325, May 2013.
Cambridge Univ. Press, 2004.
[5] N. Sun and Y. Fang, “New energy analytical results for the regulation of [27] CVX Research, Inc., CVX: Matlab Software for Disciplined Convex
underactuated overhead cranes: An end-effector motion-based approach,”
Programming, Austin, TX, USA, Apr. 2011. [Online]. Available: http://
IEEE Trans. Ind. Electron., vol. 29, no. 12, pp. 4723–4734, Dec. 2012.
cvxr.com/cvx
[6] K. Zavari, G. Pipeleers, and J. Swevers, “Gain-scheduled controller de-
[28] Gurobi Optimizer Reference Manual, Gurobi Optimization, Inc., Bad
sign: Illustration on an overhead crane,” IEEE Trans. Ind. Electron.,
Homburg, Germany, 2014.
vol. 61, no. 7, pp. 3713–3718, Jul. 2014.
[7] L. H. Lee, C. H. Huang, S. C. Ku, Z. H. Yang, and C. Y. Chang, “Efficient
visual feedback method to control a three-dimensional overhead crane,”
IEEE Trans. Ind. Electron., vol. 61, no. 8, pp. 4073–4083, Aug. 2014. Xuebo Zhang (M’12) received the B.Eng. degree in
[8] W. Singhose, D. Kim, and M. Kenison, “Input shaping control of double- automation from Tianjin University, Tianjin, China,
pendulum bridge crane oscillations,” Trans. ASME, J. Dyn. Syst. Meas. in 2002, and the Ph.D. degree in control theory and
Control, vol. 130, no. 3, pp. 034504-1–034504-7, May 2008. control engineering from Nankai University, Tianjin,
[9] W. Singhose and J. Vaughan, “Reducing vibration by digital filtering in 2011.
and input shaping,” IEEE Trans. Control Syst. Technol., vol. 19, no. 6, He is currently an Associate Professor with the
pp. 1410–1420, Nov. 2011. Institute of Robotics and Automatic Information
[10] N. Sun and Y. Fang, “An efficient online trajectory generating method System and the Tianjin Key Laboratory of Intelligent
for underactuated crane systems,” Int. J. Robust Nonlin. Control, to be Robotics, Nankai University. His research interests
published, DOI: 10.1002/rnc.2953. include mobile robotics, motion planning, visual ser-
[11] R. Liu, S. Li, and S. Ding, “Nested saturation control for overhead voing, and visual sensor network.
crane systems,” Trans. Inst. Meas. Control, vol. 34, no. 7, pp. 862–875,
Oct. 2012.
[12] Q. H. Ngo and K.-S. Hong, “Sliding-mode antisway control of an off-
shore container crane,” IEEE/ASME Trans. Mechatronics, vol. 17, no. 2,
Yongchun Fang (S’00–M’02–SM’08) received the
pp. 201–209, Apr. 2012.
B.S. degree in electrical engineering and the M.S.
[13] W. Wang, J. Yi, D. Zhao, and D. Liu, “Design of a stable sliding-
degree in control theory and applications from
mode controller for a class of second-order underactuated systems,” Proc.
Zhejiang University, Hangzhou, China, in 1996 and
Inst. Elect. Eng.—Control Theory Appl., vol. 151, no. 6, pp. 683–690,
1999, respectively, and the Ph.D. degree in electrical
Nov. 2004.
engineering from Clemson University, Clemson, SC,
[14] H. Lee, Y. Liang, and D. Segura, “A sliding-mode antiswing trajectory
USA, in 2002.
control for overhead cranes with high-speed load hoisting,” Trans. ASME,
From 2002 to 2003, he was a Postdoctoral Fellow
J. Dyn. Syst. Meas. Control, vol. 128, no. 4, pp. 842–845, Apr. 2006.
with the Sibley School of Mechanical and Aerospace
[15] D. Liu, W. Guo, and J. Yi, “Dynamics and GA-based stable control for a
Engineering, Cornell University, Ithaca, NY, USA.
class of underactuated mechanical systems,” Int. J. Control Autom. Syst.,
He is currently a Professor with the Institute of
vol. 6, no. 1, pp. 35–43, Feb. 2008.
Robotics and Automatic Information System, Nankai University, Tianjin,
[16] R. Toxqui, W. Yu, and X. Li, “Anti-swing control for overhead cranes with
China. His research interests include AFM-based nanosystems, visual servoing,
neural compensation,” in Proc. Int. Joint Conf. Neural Netw., Vancouver,
and control of underactuated systems, including overhead cranes.
BC, Canada, Jul. 2006, pp. 9447–9453.
[17] M. S. Park, D. Chwa, and S. K. Hong, “Antisway tracking control of over-
head cranes with system uncertainty and actuator nonlinearity using an
adaptive fuzzy sliding-mode control,” IEEE Trans. Ind. Electron., vol. 55,
no. 11, pp. 3972–3984, Nov. 2008. Ning Sun (S’12) received the B.S. degree in mea-
[18] T. A. Le, G.-H. Kim, M. Y. Kim, and S.-G. Lee, “Partial feedback lin- surement and control technology and instruments
earization control of overhead cranes with varying cable lengths,” Int. J. (with honors), and the B.S. degree in marketing
Precis. Eng. Manuf., vol. 13, no. 4, pp. 501–507, Apr. 2012. from Wuhan University, Wuhan, China, in 2009.
[19] Y. Fang, B. Ma, P. Wang, and X. Zhang, “A motion planning based He is currently working toward the Ph.D. degree in
adaptive control method for an underactuated crane system,” IEEE Trans. control theory and control engineering in the Insti-
Control Syst. Technol., vol. 20, no. 1, pp. 241–248, Jan. 2012. tute of Robotics and Automatic Information System,
[20] N. Sun, Y. Fang, X. Zhang, and Y. Yuan, “Transportation task-oriented Nankai University, Tianjin, China.
trajectory planning for underactuated overhead cranes using geometric His research interests include motion planning,
analysis,” IET Control Theory Appl., vol. 6, no. 10, pp. 1410–1423, nonlinear control, and control of underactuated
Jul. 2012. mechatronic/robotic systems, including cranes.
Copyright of IEEE Transactions on Industrial Electronics is the property of IEEE and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.