294 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 16, NO.

3, JUNE 2000

Hyperbolic Trajectories for Pick-and-Place

Operations to Elude Obstacles
Carmen María Müller-Karger, Andrés Leonell Granados Mirena, and José T. Scarpati López

Abstract—Trajectories of hyperbolic type have been proposed The trajectory planning may be conducted in joint or the
for pick-and-place operations where the initial and the final posi- Cartesian space. The robot motion is implemented in the
tions are known in both Cartesian and Joint spaces. The method joint-space to complete any task, thus although the initial and
only requires defining the inverse kinematics in these positions.
This process is performed with the help of a normalized hyper- final position of the robot actuator are prescribed in Cartesian
bolic trajectory, which may be a symmetric curve for the simpler space, it is necessary to know the joint variables at any instant.
case. This curve goes from zero to unity during a nondimensional The advantage of doing the trajectory planning directly in
time interval of 1 and is constructed with the composition of a hy- the joint-space is then obvious, since, if the variables to be
perbolic tangent function and another function to rescale the time
interval from [0, 1] to [ , + ]. The final function is such that
controlled are known, the task can be achieved in real time.
Conversely, when the joint variables are known, the Cartesian
the velocity, acceleration, jerk, and higher derivatives are all zero
at the extremes points. The normalized trajectory is then scaled for coordinates are easily calculated by applying a direct kine-
every variable in the joint space. Immediately, direct kinematics is matics algorithm. When the trajectory planning is done in the
applied for all those points that compose the entire discrete trajec- Cartesian-space, the process is not so simple. In this case, once
tory in joint space to generate a smooth Cartesian path. The hy- the Cartesian coordinates are established, an inverse kinematic
perbolic approach permits the definition of a third intermediate
position to avoid obstacles, which may produce a change in the algorithm must be accomplished for every position in order to
symmetry of the normalized trajectory in joint space. The inter- obtain the variables to be controlled. This is a very expensive
mediate point is tuned with the modulation of the parameters of computational process, and it cannot be achieved in real time.
the hyperbolic curve. The method was applied to a simulator of 6 Even more, the inverse kinematic algorithm does not have a
degree-of-freedom robot arm developed within this research. The unique solution.
simulator was able to avoid a specific obstacle.
Considerable effort has been invested in trajectory planning.
Index Terms—Hyperbolic trajectories, obstacle evasion, In particular, the researchers Paul [1], Taylor [2], and Luh
pick-and-place operations, robot arm. and Lin [3] proposed paths made up of straight line segments
connected together by smooth transitions with controlled
I. INTRODUCTION acceleration. Edwall et al. [4] and Lin et al. [5] utilized cubic
and quartic spline functions for constructing joint trajectories.

T HE case in study is the so-called pick-and-place operation

(PPO), that is one of the most typical trajectory-planning
techniques. PPO refers to the operation in which the robot arm
Lin and Chan used a method of joint trajectories generation
for mechanical manipulators using X-splines and quartic spline
functions. These methods require solving a system of equations
takes an object from a given initial position and puts it in a final
every time they plan a trajectory. Later on, Angeles et al.
position, specified likewise, without conferring special care to
[6] proposed an alternative approach to trajectory planning
the necessary path to achieve the objectives of the given task.
which is also based on the concept of spline functions, but in
This type of operation is mostly used in manufacturing oper-
these cases, no equation solving is required. The trajectory is
ations, in which motion is very simple and usually character-
synthesized from the scaling of a suitably normalized spline.
ized by a uniform velocity and repetitive processes. For these
Xue et al. [7] studied the path planning of two planar robot
systems, it is possible to develop a methodology to move the
manipulators that cooperate in carrying a rectangular object
robot arm along a specific geometric path avoiding obstacles.
from an initial position and orientation to a destination position
Among this type of system are the loading of belt conveyors,
and orientation in a two–dimensional (2-D) environment, but
tool changes in machine tools, and simple assembly operations
they did not address the three-dimensional (3-D) problem.
such as putting roller bearings on a shaft. An important feature
Other researchers have attacked the problem of minimizing the
in the PPO is a smooth trajectory, which is accomplished by
cost and time of moving a robot arm along a specific Cartesian
analyzing the continuity of the curve and its three subsequent
path [8]–[14], which is a different problem of a PPO.
derivatives (velocity, acceleration, and jerk), and on the other
One of first researchers who considered obstacle avoidance
hand it is essential to verify that no collision may occur.
was Lozano-Pérez [15], who mapped the obstacles in the joint
space, thus producing obstacles in the joint space in the form
Manuscript received August 28, 1998; revised June 10, 1999. This paper was of regions that the joint-space trajectory should avoid. Brooks
recommended for publication by Associate Editor A. Kochhar and Editor P. Luh [16] considered the problem of obstacle avoidance as well, by
upon evaluation of the reviewers’ comments. the introduction of the concept of freeways in both Cartesian
The authors are with the Department of Mechanical Engineering, Univerisity
Simón Bolívar, Caracas 1080 A., Venezuela. and joint coordinate space where the system is able to transit
Publisher Item Identifier S 1042-296X(00)04734-0. without danger of collision, while the difficulties entailed by the
1042–296X/00$10.00 © 2000 IEEE
MÜLLER-KARGER et al.: HYPERBOLIC TRAJECTORIES FOR PPO’S TO ELUDE OBSTACLES 295

find-path methods are avoided. Angeles [17] proposes trajecto-

ries with higher-order polynomials that allow the definition of
intermediate coordinates in the Cartesian space, these interme-
diate coordinates lie between the initial and the final position,
that are determinate in order to avoid collision. This requires
the application of inverse kinematics in these new positions to
be incorporated in the analysis.
In this work, trajectories of hyperbolic type have been
proposed for PPO’s in which the initial and the final position
are known in both Cartesian and joint spaces. The trajectory is
scaled to obtain the variable in joint-space to accomplish the
desired task in Cartesian coordinates. The hyperbolic approach
permits the definition of a third intermediate position that
lies between the initial and the final configuration, to avoid
obstacles. The intermediate point is evaded by modulating the
parameters of the hyperbolic curve. This may produce a change Fig. 1. Hyperbolic tangent function.
in the symmetry of the normalized trajectory in joint space.
Defining the inverse kinematics for this additional position is
The trajectory will satisfy the final smoothness conditions
absolutely necessary. The controlled change of the symmetry of
given by
the normalized trajectory finally produces the desired evasion
of the obstacle in the Cartesian space.

II. DESCRIPTION OF THE SYSTEM (4)

The system to be analyzed is a -joint robot manipulator. The
The solution proposed here is based on the fact that if the joint
motion of this type of system can be defined in terms of its
velocity and acceleration vectors ( and ) vanished, so will the
joint coordinates, described with a -dimensional vector , or
Cartesian velocity and acceleration ( and ). This is easily ver-
in terms of the Cartesian coordinates described with a six-di-
ified since the joint-space and the Cartesian space coordinates
mensional vector . The components of this last vector are the
are related by the Jacobian matrix. Hence, if denotes the
Cartesian coordinates of one particular point of the end effector
Jacobian matrix that relates the Cartesian and the joint veloci-
(vector ) and the Euler angles ( ) that define the orien-
ties, we have
tation.
This is

(5)
(1)

The Euler angles are calculated from the rotation matrix III. NORMALIZED HYPERBOLIC TRAJECTORY
, as indicated in the literature. The matrix is com- The normalized hyperbolic trajectory is constructed with the
posed by the unit vectors of the coordinate frame, composition of two functions. The first one is the hyperbolic
that are attached onto the end effector located at , with , , tangent that goes from to in its independent variable
and , the unit normal, unit slide, and unit approach vectors, and gives values going from 1 to 1, as shown in Fig. 1. The
respectively. This information is usually represented by a 4 point in the middle is located at , and the main variation
4 matrix as follows: is around this point in the interval where the
function changes from to The equation of this function
is as follows:

(2)
(6)

The initial and final values will be known in both Cartesian

and joint coordinates. If the time is denoted by the initial con- To change the interval to [0, 1], which is appropriate
dition (for ), and the final position for may be for a normalized trajectory, a second function to rescale the time
written as must be defined. The expression for this function is

and (3) (7)

296 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 16, NO. 3, JUNE 2000

So, the first derivative of the function could be obtained as

follows:

(13)

with

(14)

and returning to the initial variables, the following equation is

finally obtained:

(15)

From (15). the fact that the first derivative vanishes at the ex-
treme points is easy to prove. Similarly. the second-order deriva-
Fig. 2. Function x to rescale the time.
tive is obtained as

(16)

with

(17)

and once again, the second-order derivative (16) vanishes at

the extreme points. In the same way, the fact that the third-
and higher-order derivatives also vanishes at this points can be
proved easily. The factor that is always present in the
derivative formulae, tends to zero at the extreme points in the
form of an exponential function, while the factors , etc., tend
to infinity in the form of potential functions. Thus, by this rea-
Fig. 3. Comparison between normalized trajectories. soning, the indetermination zero times infinity results in zero
for all the derivatives of the hyperbolic trajectory In con-
This is represented in Fig. 2. The variable represents the clusion, (8) is not singular.
nondimensional time in the interval Finally, the com- The cycloidal trajectory has first-, second-, and third- (jerk)
position of the above-mentioned functions gives the normalized order derivatives zero at the extreme points. The sinusoidal tra-
hyperbolic trajectory, which in a compact form is expressed as jectory has only first-order derivative zero. As more derivatives
of the trajectory vanish at the extreme points, the trajectory
(8) becomes smoother near the takeoff and landing position. This
Fig. 3 shows the curve of the normalized hyperbolic trajectory property makes the hyperbolic trajectory the best, the cycloidal
compared with the normalized cycloidal trajectory trajectory the second best, and makes the sinusoidal trajectory
the worst.
(9)
IV. TRAJECTORIES AND THEIR DERIVATIVES
and the normalized sinusoidal trajectory The trajectory in the joint-space is
(10) obtained by a simple scaling of , and its argument is as
follows:
In the hyperbolic trajectory, all derivatives become zero in
the extreme points as will be shown. In order to reduce (18)
the computer calculation of the normalized hyperbolic trajec-
where and are the initial and the final positions, respec-
tory and their first-order derivatives, it is convenient to make
tively. The time will be arbitrarily defined as zero at the initial
the following change of variables:
configuration and at the final configuration. The param-
(11) eter describes the total degrees of freedom of the mechanical
system for PPO’s. This procedure permits the computation of
with the trajectories in the Cartesian space
by means of direct kinematics. Making inverse kinematics is
(12) only necessary in the initial and the final position.
MÜLLER-KARGER et al.: HYPERBOLIC TRAJECTORIES FOR PPO’S TO ELUDE OBSTACLES 297

The equation of the straight line will be constructed con-

taining the point perpendicular to and cutting it at
The point could be obtained as follows:

(21)

with

(22)

On the same basis, the equation for the straight line will
be
Fig. 4. Definition of work-space during obstacle evasion.
(23)
In the trajectory (18), there is a unique normalized trajectory
with
for all the joint variables. In the case of operations to avoid
obstacles, defining as many normalized trajectories as the
total degrees of freedom is necessary. Thus, (18) may be re- (24)
placed by
The points and are obtained by evaluating the vector
(19) function in and as shown in Fig. 4.
These are the farthest and closest points to the line , respec-
The first- and the second-order derivatives of these trajecto- tively. The next step will be to impose the trajectory to pass over
ries are calculated as a selected point between and The selection of the most
suitable point, will depend on the physical characteristics of
the workspace and also on the desired trajectory.
The objective will be to impose the end-effector trajectory to
reach a configuration in Cartesian space, with a known ve-
(20)
locity at a specific time that passes over the point
To define the space an arbitrary but reachable orientation
where and may be taken from (11), (15), and
of the manipulator is utilized. To determine the joint variables
(16).
corresponding to an inverse kinematics pro-
cedure is performed. The expressions (19) and (20), adequately
V. EVASION OF OBSTACLES handled, may be solved to obtain the values of the function
and its derivatives defined as
Avoiding an obstacle is equivalent to forcing the trajectory to
pass over a point far from the obstacle, and this is the final ob-
jective in this work. Consider a robotic manipulator that must
develop a PPO from a given initial pose specified by the po-
sition of one of the end-effector points and oriented with respect (25)
to a certain coordinate frame, to a final pose specified like-
wise. Consider that the end-effector path must evade the point Note that the configuration must lie between the initial
an arbitrary obstacle placed in the work environment. Let and final configuration of the path which imposes a limitation
be the radius of the spherical space into the work space, which on the radius of Fig. 4. To modify the hyperbolic trajectory the
must be evaded by the end effector, to ensure that any collision parameters and of (6), must be established, for
with the obstacle would be impossible. all joint variables at a simultaneous time Depending on the
Realize that the three points defined by vectors position of the obstacle, one of the next three blocks of condi-
lie in unique plane named in the work space, as tions should be selected as in (26), shown at the bottom of the
is shown in Fig. 4. Hence, the above-defined sphere becomes next page, with (I) when is near the take-off positions, (II)
a circle inside such a plane and the problem is simplified. The when is in the middle positions, and (III) when is near the
objective will be to find a point located at the perimeter of the landing positions, as shown in Fig. 5(a) and (b). Fig. 6 shows an
circle and to constrain the trajectory to pass over it without en- example of the dialog box to modify the parameter to obtain the
tering into the circle zone. desired shape of the trajectories.
The equations of the straight lines and can be defined The conditions (26) cannot be accomplished simultaneously,
with the pair of points and , respectively. The and the nondimensional time must be the same in each joint
line will represent the shortest and simplest path between trajectory; hence, an average value for the normalized trajectory
and was defined as follows:
298 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 16, NO. 3, JUNE 2000

(a) Fig. 6. Dialog box to change trajectory parameters.

The parameter represents the total degrees of freedom of

the mechanical system, and the parameter represents a weight
factor defined arbitrarily according to the influence of the joint
variable in the positioning of the end effector. The block of
conditions will be selected according this average, and in the
case that , the current block should be rejected and the
block (II) must be selected.
For each block, the values of are solved from the first
derivative of the normalized trajectory, (15), evaluated in the
current point depending on the selected condition: take off,
middle, or landing. The values for and are obtained from
(7) and (14), respectively.

VI. EXAMPLE
Consider the PUMA 560 robot manipulator whose De-
(b) navit–Hartenberg parameters are given in Table I according
Fig. 5. Obstacle near the (a) take-off position and (b) landing position.
to Angeles [17], which must accomplish the PPO defined
by the initial and the final configurations whose Cartesian
coordinates are shown in Table II. Consider besides that the end
effector must evade the obstacle whose Cartesian coordinates
are also shown in Table II, with a smooth trajectory. The
algorithm developed for the evasion of obstacles is applied
(27) with mm and the points far and near from the straight
line path are obtained, and , respectively. The Cartesian
coordinates are shown in Table II, and the joint coordinates,

(I) (II) (III)

(26)
MÜLLER-KARGER et al.: HYPERBOLIC TRAJECTORIES FOR PPO’S TO ELUDE OBSTACLES 299

TABLE I
DENAVIT–HARTENBERG PARAMETERS

TABLE II
CARTESIAN COORDINATES OF INITIAL AND FINAL POSITIONS

(a)
TABLE III
JOINT COORDINATES OF INITIAL AND FINAL POSITIONS

obtained through kinematic inversions, are shown in Table III,

respectively, for each configuration.
All the inputs that are necessary to the developed algorithm for
computingthevaluesof and thatconstrainthenormalized
trajectory to overpass a specific point are given. Once the values
and values are obtained, the trajectories for the six joint
variables are calculated. A direct kinematic algorithm produces
the Cartesian coordinates for every point of the path.
Fig. 7(a) and (b) shows the normalized trajectories for the
path in the collision case and in the obstacle-avoidance case.
(b)
Notice that for the evasion case, curves 2, 3, and 5 take off faster
than in the collision case, and curve 6 takes off slower than in the Fig. 7. (a) Normalized trajectories. (b) Normalized trajectories to evade
obstacle.
collision case, while curve 1 remains almost the same for both
cases. In the case of Fig. 7(b), all the parameters were modulated
according to the requirements to evade the obstacle. lated by simple direct kinematics. Since no kinematic inversion
A robot arm simulator, which was developed within this is required in intermediate points, the overall procedure is ap-
project [18], performed the PPO process successfully. Fig. 8(a) plicable in real time. The hyperbolic approach presents the ad-
and (b) shows the simulation of the paths in each case and in vantage over the cycloidal that the trajectories may be modified
different views. Note that the continuous line shows the path in with the modulation of the parameters It allows dif-
the case of evasion and the discontinuous line shows the path ferent speeds at taking off and landing positions, and the control
in the case of collision in all these figures. of the velgradient for each joint variable at any time. The hyper-
bolic curve also permits the modification of the parameters
in the joint space to modulate the trajectory to pass over
VII. CONCLUSION a specific point, with the objective of evading an obstacle. This
A method to generate trajectories, of hyperbolic type, for ma- intermediate position implies another kinematic inversion that
nipulators of any number of degrees of freedom has been suc- might occasion some complications because of the possibilities
cessfully developed. With a given set of Cartesian coordinates of degenerated manipulator configurations. This evasion algo-
that define the initial and the final position of a PPO, a nor- rithm needs to lie between the initial and final configurations for
malized curve is scaled throughout the entire range, for all the all of the joint variables of the intermediate position, throughout
joint variables. The procedure only involves inverse kinematics each link of the manipulator and that impose a limitation in the
in the extreme positions, and intermediate positions are calcu- selection of : the evasion radius.
300 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 16, NO. 3, JUNE 2000

[9] Y.-H. Chang, T. T. Lee, and C.-H. Liu, “On-line approximate Cartesian
path trajectory planning for robotic manipulators,” IEEE Trans. Syst.,
Man, Cybern., vol. 22, pp. 542–547, 1992.
[10] K. Shin and N. McKay, “A dynamic programming approach to trajectory
planning of robotic manipulators,” IEEE Trans. Automat. Contr., vol.
AC-31, pp. 491–511, 1986.
[11] , “Minimum-time control of robotic manipulators with geometric
path constrains,” IEEE Trans. Automat. Contr., vol. AC-30, pp. 531–541,
1985.
[12] H. H. Tan and H. H. Potts, “Minimum time trajectory planner for the
discrete dynamic robot model with dynamic constraints,” IEEE J. Robot.
Automat., vol. 4, pp. 174–185, Apr. 1988.
[13] H. H. Tan, “A discrete trajectory planner for robotic arms with six de-
grees of freedom,” IEEE J. Robot. Automat., vol. 5, pp. 681–690, Nov.
1989.
(a) [14] K. Youcef-Toumi and T. Y. Kuo, “High-speed trajectory control of a
direct-drive manipulator,” IEEE J. Robot. Automat., vol. 9, pp. 102–108,
Feb. 1993.
[15] T. Lozano-Perez, “Automatic planning of manipulator transfer move-
ment,” IEEE Trans. Syst., Man, Cybern., vol. SMC-11, pp. 681–698,
1981.
[16] R. A. Brooks, “Planning collision-free motions for pick-and-place oper-
ations,” Int. J. Robot. Res., vol. 2, no. 4, pp. 19–44, 1983.
[17] J. Angeles, Fundamentals of Robotics Mechanical Systems: Theory,
Methods, and Algorithms, ISBN 0-387-85 540-7. New York:
Springer-Verlag, 1997.
[18] C. M. Müller-Karger, J. T. Scarpati, L. E. Rodríguez, and A. Granados,
“Simulation program for 6 degree of freedom robot arm,” Boletin Téc-
nico, IMME, vol. 36, no. 3, pp. 1–17.

Carmen María Müller-Karger was born in

1966. She received the B.S. degree in engineering
mechanics from the University Simón Bolívar,
Caracas, Venezuela, in 1991, and the [Link]. degree
in toromachinery from the University of Virginia,
Charlottesville, in 1994.
(b) From 1990 to 1992, she was an Instructor with
Fig. 8. (a) Isometric and (b) top views of the simulation. the Department of Mechanical Engineering at the
University Simón Bolívar, Caracas, Venezuela. Since
1994, she has been a Professor in the Department
of Mechanical Engineering at the University Simón
The example included in the paper is the product of the exper- Bolívar. Her research interest includes the simulation of mechanical system in
the areas of robotics and biomechanics. Since January 1998, she has been a
iments done with a graphic simulator developed for the generic Researcher in the Bioengineering Center of the Central University of Venezuela
manipulator PUMA-560. The robot arm is able to pick up an ob- in the doctoral program.
ject, take it to another place, and evade an obstacle when spec-
ified. Since the simulator imitates a real robot arm, it can be Andrés Leonell Granados Mirena was born in
concluded that the process works in theory as well as in reality. 1959. He the B.S. and M.S. degrees in mechanical
engineering from the Simón Bolívar University,
Caracas, Venezuela, in 1982 and 1988, respectively.
REFERENCES From 1988 to 1992, he was with INTEVEP, a
[1] R. Paul, “Manipulator Cartesian path control,” IEEE Trans. Syst., Man, research institutes for petroleum industry, where
Cybern., vol. SMC-9, pp. 702–711, Nov./Dec. 1979. he was Leader of projects for tar and heavy crude
[2] R. H. Taylor, “Planning and execution of straight line manipulator tra- oil production, transportation, and processing.
jectories,” IBM J. Res. Develop., vol. 23, no. 4, pp. 424–436, 1979. Currently, he is a Professor with the Department
[3] J. Y. S. Luh and C. S. Lin, “Approximate joint trajectories for control of Mechanical Engineering at University Simón
of industrial robots along Cartesian paths,” IEEE Trans. Syst., Man, Cy- Bolívar, Caracas, Venezuela, and he leads the
bern., vol. SMC-14, pp. 444–450, May/June 1984. computational mechanics research group. His research interests include
[4] C. W. Edwall, C. Y. Ho, and H. J. Pottinger, “Trajectory generation and computational mechanics and primary simulation in fluid mechanics.
control of a robot arm using splines functions,” Robots VI, pp. 421–445,
1982.
[5] C. S. Lin, P. R. Chang, and J. Y. S. Luh, “Formulation and optimization José T. Scarpati López was born in Caracas,
of cubic polynomial joint trajectories for industrial robots,” IEEE Trans. Venezuela in 1973. He received the B.S. degree in
Automat. Contr., vol. AC-28, pp. 1966–1073, Dec. 1983. mechanical engineering at the University Simón
[6] J. Angeles, A. Alivizatos, and P. J. Zsombor-Murray, “The synthesis of Bolívar, Caracas, Venezuela, in 1995. He is working
smooth trajectories for pick-and-place operations,” IEEE Trans. Syst., toward the M.S. degree in automotive engineering
Man, Cybern., vol. 18, pp. 173–178, Jan./Feb. 1988. at Chalmers University of Technology, Gothenburg,
[7] Q. Xue, A. A. Maciejewski, and P. C. Sheu, “Determining the colli- Sweden.
sion-free joint space graph for two cooperating robot manipulators,” From 1995 to 1996, he was a Mechanical Engi-
IEEE Trans. Syst., Man, Cybern., vol. 23, pp. 285–293, 1993. neer at a consulting engineering firm in Caracas,
[8] Po-R. Chang and L. Chun-Shin, “Joint trajectories of mechanical ma- Venezuela. He was a Research and Teaching Assis-
nipulators for Cartesian path approximation,” IEEE Trans. Syst., Man, tant with the Department of Mechanics, University
Cybern., vol. SMC-13, pp. 1094–1102, 1983. Simón Bolívar from 1996 to 1998.