Int J Adv Manuf Technol (1992) 7:93-100

9 1992 Springer-Verlag London Limited

fldvanced
manufacturing
lechnologg
An Ultrasonic Collision Detection System
Larry J. Kutz, Ravindran Rajagopalan* and Kofi Nyamekye?
Department of Agricultural Engineering, Auburn University, Alabama, USA; *Anderson Square D Company, Leeds, Alabama
35094, USA; and tDepartment of Industrial Engineering, Auburn University, Alabama, USA

Robots in manufacturing are increasingly being called

CNC HO(~ER
on to do complex tasks that require intelligence beyond
merely following a preprogrammed path. In robotic
assembling of mechanisms, welding, machine tending IMAZAK)
and other tasks, sensing enables robots to adapt to their
environments.
In this research, an ultrasonic collision detection
system for an industrial robot was designed, constructed
and tested. Two ultrasonic transducers and ranging
modules were mounted on the robot wrist to detect and
prevent collisions with objects placed in the end effector's
path. Experiments were conducted to determine object- ['-['-'~ FINA L
CELL HOST / AT) ~.~ INSPECTI O N
to-robot distance as a function of robot speed after the
robot had stopped. Two robot motions and two loads
were studied. Statistical methods of stopping distance
vs. robot speed are presented and will be useful in
planning robot tool paths.
This uhrasonic collision detection system can be used
on stationary and mobile robots, automatic-guided D = DECOUPLER

vehicles, and other manufacturing applications.

Keywords: Robot; Sensors; Collision detection; Ultra- Fig. 1. Unmanned robotic cell at Auburn University (Shaded
sonics machines to be added).

At Auburn University, an unmanned, intelligent

robotic cell is being developed to manufacture a family
1. Introduction of gear coupling parts. The cell consists of an automatic
retrieval (and storage) system (ARS), a DoAll CNC
A robotic cell consists of a group of computer numerical 1216 band saw, two Mazak Quick Turn 10N lathes
control (CNC) machine tools arranged around a robot with Fanuc controllers, and an ASEA IRB 90/2 robot
(Fig. 1) that performs tasks such as material handling (Fig. 1). For the cell to operate autonomously, the
and machine loading and unloading in the cell. robot must sense and adapt to multiple shapes and
Typically, such cells are unmanned and must rely sizes of parts. Also, sensors on the robot and other
heavily on sensors and intelligent control systems to machine tools must detect problems in the work space
operate for long periods without human intervention. and feed this information to the control system for
diagnosis and if possible, correction during the work
cycle.
Sensor development is the key to the future of
Accepted for publication: 21 December 1990 unmanned, flexible manufacturing cells [1-3]. The
following problems that could occur during this cell's
Correspondence and offprint requests to: Larry J. Kutz, PE, Assistant
Professor, Department of Agricultural Engineering, 200 Agricultural operation show the importance of sensors in making
Engineering Buildings, Auburn University, Alabama, 36849, USA. robot control systems intelligent:
94 L. K. Kutz et al.

1. A stocking crate is left in front of the band saw (2) decide whether the robot is in danger of colliding
and is also in the robot's path. The robot moves to with the object, and (3) stop the robot if a collision is
retrieve a part from the band saw and crashes into imminent.
the crate. Noncontact range finders can measure the distance
2. A fixture or jig breaks, causing the part's location to an object. Koenigsberg [7] grouped noncontact
to deviate from the robot's preprogrammed pickup range finders into three categories based on their
point. The robot fails to retrieve the part from the operation modes: mechanical, electromechanical and
band saw. electromagnetic. Mechanical noncontact distance sen-
3. A decrease in pressure to the robot's pneumatic sors include pneumatic and acoustic sensors. An
gripper prevents the robot from grasping the part example of pneumatic sensor is a fixed nozzle through
on the band saw. which air escapes. As an object is brought into the
airstream, an increase in pressure upstream from the
In most robotic-centred cells, an important sensing orifice is developed and can be sensed by a simple
requirement is to detect obstacles in the robot's path. pressure transducer. This method is practical for
To avoid obstacles by generating alternative trajectories ranging distances of less than 1 cm and thus is not
or paths around the obstacle may also be important, useful for collision detection that requires 0.1-10 m
particularly for mobile robots. ranging distances.
Collision detection and collision avoidance methods Acoustic sensors such as sonar and ultrasound use
prevent collisions between a robot and an obstacle. sound waves. Ultrasonic transducers have distinct
The collision avoidance method - a computationally advantages as ranging devices in collision detection.
intensive problem in real-time robot control [4-6] - These compact, lightweight sensors can be mounted
determines a collision-free path for a robot moving to a robot wrist without impairing the arm's function.
among a set of obstacles. Existing collision avoidance US Patent 4 636 996 [8] describes a mobile robot that
schemes can handle only a limited range of predefined has an array of ultrasonic transducers for detecting
obstacle shapes and robot motions. objects. Working on robot collision detection, Marsh
In the collision detection method - a much simpler et al. [9] also combined information from several
problem - if an obstacle is sensed, the robot stops. In ultrasonic transducers with a priori information to
collision avoidance systems, collision detection sensors generate three-dimensional images of simple objects.
determine the presence or absence of an object in the Ultrasonic sensors are also used in navigation,
robot's operating path. positioning, imaging and simple distance measuring
Collision detection is important for several reasons [10-12]. Guichard and Renault [13] describe an ultra-
[7]: sonic ranging sensor application for adapting a robot's
1. An automatic collision detection system can prevent path to compensate for the position variations of car
a robot from crashing into an obstacle, particularly bodies moving on an assembly line as the robot sprays
when the robot is programmed off-line without them with sealant.
thorough knowledge of the cell environment. An ultrasonic ranging module developed by Polaroid
Corporation for automatic camera focusing has simpli-
2. A collision detection system can protect expensive fied and lowered the cost of noncontact distance
machine tools and devices in a manufacturing measurement [14]. On a mobile sentry robot, a ranging
environment. system designed for navigation by the US Naval Ocean
3. From the collision detection system data, a model Systems Center in San Diego, California, used the
or world map of the work cell can be built for Polaroid ultrasonic transducer with a ranging module
use in graphic simulation, collision avoidance and made by Texas Instruments (TI) [6]. This ranging
facilities layout. system had an array of five sensors for distance
The purpose of this project was to design and build measurement. It shows promise for further sensing
a collision detection system, based on an inexpensive applications in manufacturing and related industries.
ultrasonic sensor, for an A S E A IRB 90/2 robot in Electromechanical sensors, including linear and
a cellular manufacturing environment. This article rotary variable displacement transducers and magneto-
describes the hardware, control software and detailed acoustic sensors, restrict distance measurement to a
operation of the system. It also describes the testing fixed linear or rotary axis [7]. Electromagnetic sensors
and verification of the system as a function of the include optical, magnetic, capacitive, radio propagation
robot's speed and types of motion and load. and nuclear radiation sensors. Optical sensing,
especially machine vision, is a popular research area
and may be the preferred method in the future. A
basic vision system gives two-dimensional information
2. Literature Review about a scene. This information is useful in pattern
recognition and planar object location algorithms,
Collision detection systems for robots must perform but in most material-handling robot tasks, distance
three functions: (1) measure the distance to an object, information is more important [15]. Although stereo
 An Uhrasonic Collision Detection System 95

IO IBMAT
I/O INTERFACE I
BOARDUSING
INTEL8255
INPUTSOUTPUTS
I IT Ilh
-

~ ASEAROBOT 1 I |
CONTROLLER T 0Vl 12"v
RANGN
IG
MOOUL#2
E
[ vos~,. I"1 I I'['1
INPUT~

Fig. 2. Electrical schematic of ultrasonic collision detection system.

vision can be used to measure distance, machine vision Personal Computer A T (IBM AT). Figure 2 is an
systems are expensive and image analysis algorithms electrical schematic of the complete system.
are complex; therefore, other noncontact ranging
methods are more economical.
The objective of this research was to develop and 3.1 Hardware Description
test an ultrasonic collision detection system for an
A S E A IRB 90/2 robot. Work has been done in
interfacing ultrasonic transducers to robots, but little The sensor portion of the collision detection system
work has been reported on using an ultrasonic collision consists of two Polaroid ultrasonic transducers and two
detection system in a manufacturing cell. In addition, corresponding TI ranging modules. The transducers
this research sought to determine the effect of robot are mounted to an aluminium plate attached to the
motion, load and speed on the stopping distance for robot wrist mounting plate (Fig. 3). The ranging
the robot under the ultrasonic collision detection modules are mounted close to the transducers on the
system's control. robot wrist to minimise power loss to the transducers.
The ranging modules have two operation modes:
single-echo and multiple-echo [14]. The multiple-echo

3. Ultrasonic Collision Detection

System Description and Operation

The robot ultrasonic collision system developed in this

project uses the TI SN28827 ultrasonic ranging model.
The module drives the instrument-grade, 50 kHz,
300 V Polaroid ultrasonic transducer that functions as
a speaker in the transmit mode and its a microphone
in the receive mode. The transducer has about a 14~
beam width and is extremely directional [14]. The
ranging module and transducer system can detect
objects in the 0.4-10.7 m ranging distance set in the
control software. In distance measurement, the ranging
module is accurate to -+2%.
The collision detection system developed for the
A S E A IRB 90/2 industrial robot was controlled by an Fig. 3. Ultrasonic transducers (A) and ranging modules (B) mounted
assembly language program operating on an lBM on the robot.
96 L. K. Kutz et al.

mode can be used to distinguish multiple objects at

different distances. The single-echo mode is used when
a single ranging value is desired, and was used
exclusively in this research.
A distance measuring cycle for the ranging modules INIT l L_
and transducers are initiated by switching the input
line, INIT, from a logic 0 to a logic 1 state. This action ~~- . ~ II TM'~'-~E~
(INTERNAL)
puts the ranging modules in the transmit mode, and
the ranging modules then generate 16 cycles of a
300 V, 49.4 kHz signal. At the transducers, this signal INTERNAL /-L7s
is converted to electrostatic pulses that travel toward, BLANKING

then reflect from, a target, The ranging modules ECHO [ I__

then automatically wait 2.38 ms before switching the
transducers to the receive mode, to prevent any ringing
in the transducers from being detected as echoes.
Multiplying this 2.38 ms internal blanking by the speed Fig. 4. Typical timing diagram of the electrical signals to and from
of sound gives the modules' minimum ranging distance the ranging modules.
of 0.4 m at standard calibration conditions (1 atm,
70~ able I/O board in the IBM AT (Fig. 2). The E C H O
The elapsed time to echo detection can be used to line from each ranging module was connected to
compute the distance to the target as follows. First, individual input ports of the I/O board, and to the
the speed of sound in air is calculated by [6, 16, 17] robot controller through an auxiliary circuit box. This
box has three main components:
c = x/gck R r (1)
1. An OR gate (IC 7432) that signals the robot to
where c = speed of sound in air, m/s stop if either or both of the ultrasonic transducers
gc = gravitational constant = 1 (kg m)/(N s2) detect an echo from an obstacle
k = ratio of specific heats = 1.4 2. A 30 V open collector output (IC 7407) that switches
R = gas constant = 287.08 J/(kg ~ a 24 V d.c. signal to activate the P R O G R A M STOP
T = absolute gas temperature, ~ function on the robot when the OR gate's output
Distance travelled is then simply is in a logic 1 state
3. A 5 V d.c. 6 A regulated power supply that provides
d=ct (2) 5 V to the ranging modules via the Vc,, connection
where d = distance to the target, m Activating the P R O G R A M STOP function on the
c = speed of sound in air, m/s robot controller interrupts the current robot instruction
t = one-half the pulse's round-trip time to and and stops the robot. This procedure is equivalent to
from the target pushing the robot controller's P R O G R A M STOP
In this research, Eqs (1) and (2) were used to calculate button [19].
the time interval necessary to hold the INIT line high
to detect an object in a user-defined range.
The distance measuring cycle is halted by switching
3.2 Software Description
the INIT line to a logic 0 state, which also switches
the E C H O line to logic 0. Figure 4 is a typical timing An assembly language control program in an IBM AT
diagram of the INIT and E C H O lines' states as controlled operation of the 8255 I/O board and the
functions of time. ranging modules. The program was written with
The ranging modules are interfaced to an IBM AT Microsoft Assembler version 4.0, and consists of a
through an eight-pin connector (Fig. 2). The collision main program and two subroutines. Figure 5 is a
detection system uses four of these pin-outs [18]. flowchart of the program. Note that the logic for
sensing whether an object is detected is duplicated by
1. The input line, INIT, that activates distance measur- the OR-gate (IC 7432) in the auxiliary circuit box.
ing cycle The actual stopping of the robot is done through the
2. The output line, E C H O , that signals echo detection auxiliary circuit box.
3. The power supply line, V~c The main program initialises variables, registers and
resets the interrupt rate of the IBM AT's internal
4. The ground pin, OV
system clock (IC 8254) from 18.2 Hz to the frequency
The INIT line from each ranging module was connected corresponding to the value of TMR_CNT1 [18, 20].
to individual output ports of the Harris 8255 programm- The variables TMR_CNT1 and TMR_CNT2 control
 An Ultrasonic Collision Detection System 97

-- ~ Fig- 5 (1) ( ~
Dectare code, data I M~n Program Push conlents ol r
1
Fig. 5 (2) Output a character
SUlck segrne~ts I ir~s~ctions or~ s~ac~ SERVICE Subrouline it: the keybd buffer
I I I
O e C ~ ve~ab~esand Read flag Print "Object delec~ed!"
sel TMR_CNT1, TMR_CNT2

Set new IntemJpt tale y

for IC, 8254 .
I
l Set [NIT lines low
Conllgute 8255 I,'0 I 1or TMR CNTt
board [ Sel INIT Iino$ high i
I for TMR_CNT2
Set flag = 1
Clelu keCod buffer I [
I Set flag. 0

I
r
Reset IC 8254 to
I
Call ROBOT subroutine
Ca~ default intBrPJpt
I rate _lI

Fig. 5. Flowchart of assembly language control program for the ultrasonic collision detection system: (a) Main Program;
(b) SERVICE subroutine; (c) ROBOT subroutine.

the time the INIT line is held at a logic 0 and logic 1 For maximum reflection, one face of the prism was
state, respectively. The main program then monitors placed perpendicular to the path of the ultrasonic
the keyboard for a key strike, enabling the user to signals in the horizontal and vertical planes.
exit the program gracefully, and enters a loop in which The collision detection system was tested with the
the S E R V I C E subroutine is called. A S E A robot for two robot paths and two end
The S E R V I C E subroutine controls the ranging effector loads. In one robot path, the robot arm was
modules by setting the INIT lines to either a logic 0 programmed to swing about the base - referred to as
or 1 state, depending on the flag variable's value and swinging motion (S). In the other robot path, the robot
the status of the E C H O lines. If an echo was not end effector was programmed to move radially from
detected in the time identified by T M R _ C N T 2 , then the base - referred to as linear motion (L).
no object was detected in the user-defined distance For all tests, the robot end effector was a two-jaw
and the control loops to the main program. pneumatic gripper. The transducer was mounted on
The R O B O T subroutine is called when an echo is the wrist mounting plate, 480 mm from the tips of the
detected. This subroutine simply prints a message on gripper jaws. The ranging distances set in the assembly
the console and ends the program. It could also send control program were reported as the distance from
a signal to a cell host computer, activate an alarm, or the tips of the gripper jaws to the object. The ranging
initiate a collision avoidance scheme. modules' total ranging distance was the 480 mm
transducer-to-gripper distance plus the gripper-to-
object distance. The programmed ranging distance was
4. Experimental Procedure 239 mm beyond the gripper tips. The robot end
effector's starting position was outside the programmed
The objective of the experimental tests was to verify ranging distance, and its programmed final position
the collision detection system's operation and to was beyond the obstacle. Programmed point-to-point
evaluate its performance as a function of programmed robot speeds were 50 to 700 mm/s.
robot speed. The robot was programmed in the A S E A The two loads were 0 and 16.8 kg. Therefore, four
robot language ( A R L A ) to move toward an obstacle different treatments, each consisting of five replications,
using a rectangular, wrist-oriented coordinate system. were analysed. The treatment label abbreviations rep-
That is, the robot end effector moved using straight resent the load and the robot motion; for example, 16L
line motion between programmed points [21]. The represents a 16.8 kg load and a linear motion.
obstacle was a 4 ram-thick cardboard triangular prism.
98 L. K. Kutz et al.

300,
in each case at e~ = 0.001 or smaller. All four regression
-4 2
Y = 227.3 - 0,4032X+ 1.352 x 10 X
models correlated well with the data; the correlation
R2= 0.9835 coefficients were 0.98 or better for each model.
E Further statistical analyses were done using the
E
G L M homogeneity-of-slopes model to evaluate the
6 hypothesis that the treatments had no effect on
0e- 200
stopping distance. Analysis of covariance revealed that
treatments 0S and 16S were not significantly different
C3 at o~ = 0.05. However, treatments 0L and 16L
were different from each other and were individually
.E loo different from treatments 0S and 16S. Figure 7 shows
D..
Q. plots of the four regression models as three separate
O
zx groups. Therefore, the type of motion has a significant
effect, as expected, because the robot's dynamics are
different for the robot swinging about the base than
0 200 400 600 800 for the robot arm moving radially from the base. The
load also affected the robot's stopping distance,
Robot Speed, mm/s supporting the hypothesis that the robot's increased
momentum at faster robot speeds also decreased the
Fig. 6. Robot stopping distance (Y) versus robot speed (X) for final measured distance between the gripper tips and
treatment OL (0 kg load, linear motion). the object.

5. Analysis of Results

The ultrasonic collision detection system stopped the 800

robot from crashing into the obstacle at programmed
robot speeds up to 700 mm/s. At faster speeds, the l=
response times of the collision detection system and i::
~.. ~6L
robot controller were the limiting factors. 6
Figure 6 shows the results for the 0 kg load and 0200
e-
linear motion (OL) treatment, and is typical of the m \..~ ..
\-'x "'"
other three treatments. The data points represent the ._
", \ ..
average stopping distances of five replications at each Q \.x, '. .

robot speed. The stopping distance is the final measured OS " ' - ~ \ % \ ,...
.c_ 100
distance between the gripper tips and the object. The ~.
regression line was obtained by analysis of variance e~
0 16S " < ' ~ . .
using the SAS generalised linear model (GLM) pro-
cedure [22]. A quadratic model was fitted to the data
for each of the four treatments. Quadratic models for 0 r I I I I / =
all four treatments are displayed in Table 1. 0 200 400 600 800
Analysis of variance on each of the regression Robot Speed, mm/s
coefficients for each of the four models showed that
Fig. 7. Combined plot of robot stopping distance (Y) versus robot
the linear term was significant at e~ = 0.0001. The speed (X). comparing the linear regression models for the four
quadratic term coefficient, although less important in treatments (0 and 1 6 : 0 and 16.8 kg loads: L and S: linear and
reducing the error sum of squares, was also significant swinging motions).

Table I. Quadratic models fitted to data from each of the four treatments.

Treatment Quadratic model" Correlation

coefficient
Label Load Robot
(kg) motion

0L 0 Linear Y = 227.3 - 0.4(132X + 1.352 x 10 4A~ 0,9835

0S 0 Swinging Y = 226.9- 0.4802X+ 2.526 x 10 4Xe 0.9914
16L 16.8 Linear Y = 235.1 0.3659X + 5.791 x 10 4X2 0,9969
16S 16.8 Swinging Y = 231.1 - 0,4713X + 2.278 x 10 4X2 0,9951

~'Y = stopping distance, m m : X = robot speed, mm/s.

 An Ultrasonic Collision Detection System 99

Therefore, the robot's dynamics are important in Acknowledgement

the ultrasonic collision detection system's operation.
The results show that the programmed ranging distance The authors thank the Advanced Manufacturing Tech-
must be increased as the robot speed increases so the nology Center (especially H. L. Fellows) of Auburn
controller has enough time to stop the robot before a University for its support, and A. H. Honnell of the
collision occurs. The ranging stopping distance could AMTC Information Resources Laboratory for editorial
be increased by revising the control software to assistance,
compensate automatically for robot speed during robot
operation.
The collision detection system tested in this research
will stop the robot only if an obstacle is directly in
References
front of its gripper. Objects that are not perpendicular
1. Y. Koren, Computer Control of Manafacturmg Systems,
to the ultrasonic transducer may not be sensed because McGraw-Hill, New York, 1983.
the transducer is extremely directional [14]. One 2. L. Rossol, ~'Technological barriers in robotics: a perspective
solution would be to broaden the sensing window by from industry", Robotics Research: Proceedings of the First
mounting several transducers at strategic locations International Symposium, Brctton Woods, New Hampshire, The
along the robot arm. The I/O board in this study could MIT Press, Cambridge, MA, pp. 963-972, 1984.
3. A. Pugh, "'Robot sensors - a personal view", Proceedings of
handle up to 12 ultrasonic units. Additional transducers the 1985 International Conference on Advanced Robotics,
would require more I/O channels or multiplexing Tokyo, Japan, IFS Publications Limited and North Holland/
equipment, possibly increasing the control system's Elsevier Scientific Publishers, pp. 521-532, 1985.
response time in stopping the robot. 4. O. Khatib, "'Real-time obstacle avoidance for manipulators and
mobile robots", Proceedings of the IEEE 1985 International
Conference on Robotics and Automation, St Louis, Missouri,
IEEE Computer Society Press, pp. 500-505, 1985.
5. S. Cameron, "A study of the clash detection problem in
6. Conclusions robotics", Proceedings o f the 1985 IEEE International Conference
on Robotics and Automation. St Louis, Missouri, IEEE Com-
An ultrasonic collision detection system for an ASEA puter Society Press, pp. 488--493, 1985.
6. H. R. Everett, "A multi-element ultrasonic ranging array",
IRB 90/2 robot was designed, constructed and tested. Robotics Age, pp. 13-20, July 1985.
The system used two Polaroid ultrasonic transducers 7. W. D. Koenigsberg. "'Non-contact distance sensor technology",
and two Texas Instruments ultrasonic ranging modules Intelligent Robots: Third International Conference on Robot
mounted on the robot wrist mounting plate. The Vision attd Sensory Controls (RoViSe6\~), Cambridge, MA,
SPIE, Bellingham, W A , pp. 519-528, 1984.
system was controlled by an assembly language program 8 ."Ultrasonic obstacle location apparatus and method", US Patent
running on an IBM AT and was connected to the 4 636 996, Jan. 13, 1987 (1(1), US Patent Office, Washington,
robot controller to stop the robot when an object was DC.
sensed within the programmed distance. 9. K. A. Marsh, J. M Richardson, J. S. Schoenwald and J. F.
Martin, "Acoustic imaging on robotics using a small set of
The system was tested with the robot as a function transducers", Proceedings of the 4th International Conference on
of robot speed and two types of motion and toad. In Robot Vision and Sensory Comrots (RoViSeC4), London, UK,
all tests, the programmed stopping distance was IFS Publications, Bedford, pp. 261-268, 1984.
239 mm beyond the gripper tips. The following con- 10. S. C. Pomeroy, H. J. Dixon, M. D. Wybrow and J. A. G.
Knight, "Ultrasonic distance measuring and imaging systems for
clusions were reached: industrial robots", Robot Sensors: Vol. 2 - Tactile and Non-
vision (International Trends in Manufacturing Technology), ed.
1. The ultrasonic collision system stopped the robot A. Pugh, Springer-Verlag, New York, pp. 261-270, 1986.
at robot speeds up to 700 mm/s. At faster robot 11. T, Arai and E. Nakano, "'Development of measuring equipment
speeds, the response times of the collision detection for location and direction (MELOD[) using ultrasonic waves",
system and robot controller were not fast enough Trans. ASME, 105, pp. 152-156, September 1983.
12. "Ultrasonic apparatus for positioning a robot hand", US Patent
to stop the robot before it collided with an obstacle. 4 718 023, Jan. 5, 1988 (121, US Patent Office, Washington
2. The robot's dynamics greatly affect the performance DC.
of the collision detection system in stopping the 13. M. Guichard and A. Renault, "'Industrial use of ultrasonic
ranging sensors in robotics", Robot Vision and Sensory Controls
robot. Tests showed that the types of motion and (RoViSeC6) International Conference, Paris, France, [FS Publi-
load are apparently significant factors in predicting cations, Bedford, pp. 157-164, 1986,
the stopping distance of the robot under the collision 14. S. Ciarcia, "An ultrasonic ranging system", BYTE, 9(11), pp.
detection system's control. 112-123, 1984.
15. E. Abele, U. Ahrens and G. Drunk, "'Application of ultrasonic
3. Using the ultrasonic collision detection system is an sensors for handling taks with industrial robots ~', IFS ISIR
inexpensive and reliable method for enabling a Industrial Robots 16th International Symposium, Brussels,
robot in a manufacturing cell to sense distance to Belgium, pp. 469-480, 1986.
16. D. Ensminger, Ultrasonics: Fundamentals, Technology, Appli-
an object. However, multiple ultrasonic sensor cations, Marcel-Dekker, New York, 1988.
systems are required to compensate for the trans- 17. V. M. Faires and C. M. Simmang, Thermodynamics, MacMillan,
ducer's directionality. New York, 1978.
100 L. K. Kutz et al.

18. R. Rajagopalan, "'An ultrasonic robot collision detection sys- 8086/8088 Family: Architecture, Programming and Design,
tem", unpublished M.S. thesis, Auburn University, Graduate Prentice-Hail, Englewood Cliffs, 1986.
School, Auburn University, Alabama, 1990. 21. R. P. Paul, Robot Manipulators: Mathematics. Programming
19. Robot Programming Manual CK 09-1401E, ABB Robotics Inc., and Control, MIT Press, Cambridge, MA, 198l.
16250 W. Glendale Drive, New Berlin, Wisconsin, 1984. 22. SAS Institute Inc., SAS User's Guide: Statistics, Version 6.03
20. Y.-C. Liu and G. A. Gibson, Microcomputer Systems: The Edition, SAS Institute Inc., Cary, NC, 1988.