Il1Williillilljlnllllu: '

1'"
r

[‘VQ<
.I'N.‘

This is to certify that the

thesis entitled

AN AUTOMATIC STEERING CONTROLLER FOR
A ROBOTIC LAWN TRACTOR

presented by

Ronald Earl Squires

has been accepted towards fulfillment
of the requirements for

Wegmein A°E°T°

fl 2 /
C/flM/fcczm C /ZL flog/004

Major professor

 

Date ////3 /6’/

0-7 639

 

 

 

MSU

 

 

RETURNING MATERIALS:
ace 1n oo rop to
remove this checkout from

 

LIBR
1...;352E:_ your record. FINES will
be charged if book 15
returned after the date
stamped below.
.“I s _
x 15'2”” if 'z/
1/" I f

V

042617

I!“ (7* q
Lt'd -' 1 23!

(.37

 

_..__—_————-~

 

AN AUTOMATIC STEERING CONTROLLER FOR
A ROBOTIC LAWN TRACTOR

Ronald Earl Squires

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Agricultural Engineering

1981

ABSTRACT

AN AUTOMATIC STEERING CONTROLLER FOR
A ROBOTIC LAWN TRACTOR

By
Ronald Earl Squires

As a first step towards the development of a robotic
agricultural vehicle, an automatic steering system was
developed for a robotic lawn tractor. The steering
controller guided the tractor around an ever decreasing
rectangular area to mow a lawn without operator assistance.
The microprocessor-based controller utilized no sensor or
device located separate from the tractor. The micro-
processor "learned" the size'of the rectangle as the
tractor mowed the perimeter. The controller guided the
tractor along one side of the rectangle, performed a 270
degree turn, then continued with the adjacent side. The
controller used information as to the location of the
grass cut edge, the position of the steering wheels, and
the distance traveled to determine when a steering
algorithm needed to be implemented. The robotic tractor
mowed all of a rectangular lawn automatically except for

the outside cut and the final swath.

ACKNOWLEDGEMENTS

I would like to thank Dr. Truman Surbrook for his
help and encouragement in the development of the
controller and in the writing of this thesis. I would
also like to thank Dr. John Gerrish and Dr. Michael
Shanblatt for their input to the development of the
controller. The support of the Detroit Edison Company,
both financial and in locating and repairing the lawn
tractor was vital in this research. Finally, I would
like to thank my wife, Kathy, and my son, Steve, for
their understanding, encouragement and assistance

throughout my graduate study.

ii

LIST OF
LIST OF

Chapter
I.

VI.
VII.

<.
H
H
P!
O

H
4

TABLE

TABLES . . . . . .
FIGURES . . . . .

INTRODUCTION . . .
OBJECTIVES . . . .
LITERATURE REVIEW .
SYSTEM GENERAL DESCR
SYSTEM DEVELOPMENT -

Linear Actuator .
Initial Control Bo

Microprocessor Control .

Distance Traveled
Grass Edge Sensors
Drive Relay . .

Operator Controlled Inputs

SYSTEM DEVELOPMENT -
RESULTS AND DISCUSSI

Grass Edge Sensor
Distance Traveled

Wheel Position Sensor . . .

Microprocessor .
CONCLUSION . . . .
LIST OF REFERENCES

OF CONTENTS

IPTION . .

HARDWARE AND

X 0 0

Input .

O O O

O Q C O O O
O O O O O O 0

SOFTWARE .
ON . . . .

O O O 0 O

Circuitry

O O O O
O O O O O 0

iii

MECHANICAL

iv

11
18
19
26

4

32
40
43
L9
71
80
83
84
85
88

9O

{N \A) N
0

LIST OF TABLES

Input Port Assignments .

Output Port Assignments

Register Allocation

Memory Allocation

iv

44
45
50
52

12.
13.
14.
15.
16.

17.
18.

19.

LIST OF FIGURES

Positioning of Major System Components upon
the TraCtor O O O O O O O “O O O O O O O 0

Typical Pattern Cut by Tractor in Automatic
Linear Actuator Mounting . . . . . . . . .
Linear Actuator Control Circuit . . . . . .
Initial Control Box Schematic . . . . . . .
RCA COSMAC VIP Simplified Block Diagram . .
Front Wheel Position Decoding Circuit . . .
Distance Traveled Circuit . . . . . . . . .
Grass Edge Averaging Circuit . . . . . . .
Drive Relay Circuit . . . . . . . . . . . .
Rotary Switch Schematic . . . . . . . . .

System Flow Chart . . . . . . . . . . . . .
Initialization Flow Chart . . . . . . . . .
Main Loop Flow Chart . . . . . . . . . . .
Turn Selection Flow Chart . . . . . . . . .

Correction Right Flow Chart
Correction Left Flow Chart . . . . . . . .

Corner Turn Flow Chart . . . . . . . . . .
Shutdown Flow Chart . . . . . . . . . . . .

Learn Flow Chart . . . . . . . . . . . . .

13
15
21
23
27
28
3O
33
38
42
47
54
56
57

if
1'}

61
63
66
67

CHAPTER I

INTRODUCTION

Safety, machine efficiency and energy efficiency are
important factors that can be maximized by use of robotic
agricultural equipment. A decrease in the availability
of trained agricultural equipment operators, combined with
the increase in energy and machinery costs may shift the
industry towards robotics as a way of lowering operating
costs while decreasing the demand for highly trained
operators. The development of a practical, yet versatile,
robotic agricultural machine was well beyond the scope of
this project. This project was, however, viewed as an
initial step towards the final goal of a robotic
agricultural machine.

One of the functions that a robotic agricultural
machine must be able to do is to move about the field
following an efficient pattern. The machine must be
capable of following the desired path within the accuracy
specifications of the task being performed. As dead
reckoning has been attempted and proved unsatisfactory
for all but the simplest of jobs (Gilmour 1960), the
robotic machine must be capable of performing steering
functions to maintain the desired work pattern. This

1

work, therefore, concentrated upon the development of an
automatic steering system.

At the start of the project, it was decided that for
a unit to be practical it must be capable of obtaining all
necessary information without the use of external devices.
Several types of automatic steering systems have been
utilized by various researchers in an attempt to make a
vehicle steer itself. These major types are documented
in the literature review, however, only the ones that
utilize physical parameters of the field itself were
capable of meeting the requirement of no external devices.

Once the method of primary location sensing had been
determined, a specific machine and function were required.
Several types were considered. A lawn tractor was chosen
because the space requirements for the lawn tractor were
less than that of a conventional agricultural tractor.
Also, the tractor could be tested on campus, the time of
year during which the tractor could be tested was quite
large, and finally, the cost of acquiring and developing
the system would be less when compared with a conventional
agricultural machine. It was felt that all information
obtained from this lawn tractor could be easily transferred
to a conventional farm tractor when needed.

With the help of a local electrical power supplier,
a General Electric lawn tractor was acquired. This vehicle

was totally electric. An electric vehicle was not required

for this project, however it was determined that it would
be an asset in future energy related studies.

The ability to mow a given lawn of any size and shape
would be the final goal of the project. This type of
versatility would require sophisticated robotics. There-
fore, the development was divided into a series of smaller
steps of which this thesis reports the first. Detailed in
the following sections of this thesis is the description
of the lawn tractor automatic steering controller that was

designed.

CHAPTER II

OBJECTIVES

The general objective of the robotic lawn tractor
project was to build and test a lawn tractor which would
mow a lawn without an operator in direct control. For
the initial development covered in this thesis, the lawn
was assumed to be rectangular, of uniform height and
density, and free of obstructions. The lawn tractor would
mow the lawn by following one side of the rectangle,
performing a 270 degree turn, reacquiring the grass cut
edge and then mowing the adjacent side. Repeating these
steps took the tractor around the rectangle until one
dimension of the uncut area had been reduced to less than
one meter. At that time, the controller would direct the
tractor to stop, shutdown the mower, and await the return
of the operator.

The specific objectives of this portion of the project
were:

1. The tractor must follow the grass cut edge such that
a minimum of 85 percent of the cutting width of the
mower is utilized and a minimum of 99 percent of the
total area attempted is mowed;

2. On dry grass the controller must execute a 270 degree

turn at the required time, in such a way that the
adjacent grass edge could be reacquired in less than

one meter of travel along that edge;

The microprocessor must learn the dimensions of the
initial lawn and perform the necessary mathematical
operations required to reduce the learned dimensions
such that they represent the remaining uncut portion
of the lawn, as the lawn is mowed:

The controller must halt tractor forward travel and
disable the mower whenever any dimension of the uncut
lawn is reduced to less than one meter:

The operator interface with the controller must be
such that a person unfamiliar with the controller
can operate the system after less than five minutes
of instruction;

All systems of the lawn tractor must also be operable
in the manual mode.

CHAPTER III

LITERATURE REVIEW

The desire to eliminate the operator from the tedium
of steering an agricultural vehicle has resulted in the
development of several experimental controllers. Grovum
and Zoerb (1969) classified guidance systems into three
basic types; semi-automatic, automatic command and
automatic preset. The semi-automatic category included
radio controlled vehicles and also master-slave arrange-
ments. The semi-automatic type of controller was capable
of removing the operator from a hazardous location but
did nothing to lessen the burden upon the operator. The
automatic command group utilized external systems to
monitor the position of the vehicle and issued instructions
to the vehicle to cause it to follow a desired course. A
major disadvantage of the command type of guidance was the
required high degree of component accuracy and the
resultant high cost. The last category, automatic preset,
was where the vehicle had the necessary equipment to
evaluate sensed data and to perform steering functions
based upon the data. This category of guidance systems
has received the most study.

Gilmour (1960) utilized a dead reckoning system that

proved to be impractical due to the effects of slippage
and other mechanical errors. The two most common systems
that have been studied involve the vehicle following a
physical parameter. The first involved a current-carrying
conductor buried in the ground. The vehicle followed the
conductor by observing the strength and orientation of
the produced magnetic field. Rushing (1971) described an
antenna that was capable of sensing the buried wire and of
providing information as to the orientation of the wire
with respect to the antenna. A considerable amount of
work utilizing the buried wire technology has been
performed at the National Tillage Machine Laboratory
located at Auburn, Alabama by R. E. Young and R. L.
Schafer (1974). Young, et al., (1980) described the use
of an Intel 8085 microprocessor to replace the digital
logic controller used by Schafer and Young in earlier
experiments. The change to a microprocessor was made to
improve flexibility and expand the system's capabilities.
This technique was capable of giving good results. In
fact, a lawn mower is commercially available from a
company in Alpharetta, Georgia that utilized the buried
wire technique (Marley 1981). However, the use of this
type of controller is limited to only those areas that
have been fitted with the wire network. Schafer and
Young (1978) state, "We believe that the development of

any practical automatic guidance system must include a

consideration of the elimination of the buried wire".

The second system utilized a physical property of the
field itself to provide guidance information. This type
might use a furrow in the ground or the crop itself as the
source of information. A furrow following system was built
and tested by Kirk, et al., (1976). This system utilized
both the on-off and the proportional type of steering
control. The two systems provided adequate tracking,
however the prOportional system had a slightly lower
tracking error and was more stable at higher speeds. Kirk
and Krause (1975) utilized the edge of the uncut crop to
guide a swather. The developed unit worked under normal
conditions, but it was subject to oscillations that caused
it to miss crop afterrepeated passes. These oscillations
were partially attributed to the 'play' in the swather
steering mechanism. A combine was also fitted with sensors
that allowed it to follow a corn row (Busse, et al., 1977).
These sensors were such that if the corn row was too far
towards one side, the corn would push against a mechanical
arm, indicating the relative position of the corn row.

The problems encountered with this system were as follows:
(1) The responsiveness of the steering system, (2) the
sensitivity of the controller to row skips, (3) the
inability of the sensing system to be applied to grain
crops and (4) the system would not permit easy backing up.

Upchurch, et al., (1980) designed and built an automatic

steering controller for an over the row apple harvester.
This controller sensed the trees by means of mechanical
arms contacting the trees. This controller utilized five
different types of steering; automatic, front only, four
way, crab, and rear only. An RCA 1802 microprocessor was
capable of performing all five steering modes. Accuracy
of the system was considered adequate. A commercial form
of this type of controller, utilizing the furrow following
technique, is available from a company in Modesto,
California (Baxter 1981).

It should be noted that all the systems so far
described utilize only one input in determining the
required response of the steering. Ambler, Harries and
Cox (1980) utilized two different input forms. However,
only one was utilized at any given instant in time. An
optical furrow following system was used to guide the
tractor across the field and an optical ranging method was
used to turn corners. The-ranging sensor required that
special reflective posts be placed in a line about 90
degrees to the direction of the furrow and about 15 meters
apart. This system for turning resulted in 80 percent of
the turns ending less than 30 centimeters from the desired
point. The ranger used a microprocessor to evaluate the
location of the tractor near the end of the field and
throughout the turn.

Kanetoh (1976) described a driverless combine that

10

had six automatic functions. This combine used a
mechanical arm to sense the position of the grain edge,
which controlled the direction of steering. Another
mechanical arm sensed the presence of grain and started a
turn routine when grain had been lost for a preset distance
of travel. If no grain was found after the completion of
a turn, the machine stopped forward travel. All six
automatic functions operated in isolation from each other,
except the turn and the stop controllers shared the input
of the plant stalk sensor, and the automatic steering
controller was disabled during a turn. This system worked
on a rectangular field and used a pattern that took the
machine around the ever decreasing rectangle. Turning was
done by rotating about 45 degrees to the left, backing up
and then turning another 45 degrees.

Utilization of microprocessor-based systems for
automatic steering controllers has resulted in more
versatile controllers. Microprocessor-based systems can
incorporate self-diagnostic features and can respond to a
variety of inputs to determine the best type and amount of
correction. Young, et al., (1980) reported the development
of a steering algorithm to allow for the controller to
adjust the system output to compensate for changes in
ground speed. When the automatic steering controllers are
provided with more relevant information, the quality of

control is sure to improve.

CHAPTER IV

SYSTEM GENERAL DESCRIPTION

The robotic lawn tractor was a General Electric lawn
tractor fitted with a microprocessor-based control system.
The control system utilized an RCA 1802 as the central
processor unit. Sensors and conditioning circuits provided
the microprocessor with digital inputs that represented the
status of the two grass edge sensors, the position of the
steering wheels and the completion of another unit of
distance traveled. The microprocessor controlled the
tractor by producing output voltage levels that initiated
changes in the tractor's mechanical steering system and
electrical power system.

The Detroit Edison Company provided the General
Electric lawn tractor and funded necessary repairs. The
General Electric lawn tractor had a 36 volt battery system
as the source of energy. A 36 volt direct current variable
speed motor combined with a four speed mechanical trans-
mission provided a range of tractor speeds. The front
mounted mower had three blades, each powered with a
separate 36 volt direct current motor. The mower had a
maximum cut width of 1.07 meters.

The ability of the tractor to turn sharply was

11

12

typical of this size of lawn tractor. The minimum radius
was 1.5 meters. Control of the mechanical steering system
was obtained by means of a linear actuator. When the
tractor was moving on dry grass, one second was required to
move the steering wheels from straight to a maximum turn.

Based upon limitations of the tractor, a set of
project objectives were developed. The restraints upon
the lawn were not changed- They were that the rectangular
lawn was uniform, free of obstacles and no larger than 38
meters on any one side. The restraints were placed upon
the lawn to allow the effort to be directed toward the
steering controller itself rather than the development of
sensors to handle thin or eneven lawn, obstacles or other
special problems. These problems must be investigated
before a practical lawn tractor can be built.

The location of the major controller components with
respect to the tractor is shown in Figure 1. Three
alternatives for the mowing pattern were considered. The
first required the tractor to go around the rectangle
using a 90 degree turn at the corners. The mounting
position of the mower combined with the mobility of the
tractor made the use of the 90 degree turn impractical.
The second alternative required the tractor to start at
one edge of the lawn and work across the lawn doing 180
degree turns on both ends. This pattern is referred to

as a switchback pattern- The switchback pattern was not

13

ho¢0893 map cog: masocomaoo sopmzm home: we MGH=OHufimom

 

w mmonm

 

momzmm Homm mm<mo

moe<=eo< m<mqu

momzmm onBHwom gammz qummmem
AmmHe OB ammo<ee<v wsmzo<z

Amz<mh OB ammo<aa<v moeHzm Qmmm
mw<qmm 02Hmmmem .Bomqmm A<Dz<:\oes<
maomazoo A<Dz<z

mqomazoo mo<hmmBzH moe<mmmo
mommmoommomOHz

 

14

used because sensors to locate the grass cut edge would

be required on both sides of the mower. The third
alternative required the tractor to go around the rectangle
using a 270 degree turn at the corners. This pattern
appeared to be most acceptable.

Once the 270 degree turn was selected, two patterns
of travel were examined. The tractor could start in the
center and work out, or it could start on the outside and
work inward. As the exact center of a lawn may not be as
well defined, it was felt that starting on the outside and
working inward would be the more reasonable choice. A
typical pattern is shown in Figure 2.

For maximum versatility, the size of the lawn to be
cut should be variable. To get the dimensions of the lawn
to be cut into the memory, the controller had to learn the
size of the lawn. A grass cut edge was required for the
controller to function in the automatic mode. This was
provided by having the operator mow around the outside of
the lawn. As the operator mowed the final two sides of
the rectangular lawn, the controller determined the
dimensions of the lawn to be cut. The operator used a
rotary switch to inform the microprocessor when the start
and end of a side to be learned occurred.

As the lawn tractor approached the completion of the
lawn, it is possible that the final swath could be much

less than the cut width of the mower. In fact, the final

15

 

START ST
AUT IC MANUAL

FIGURE 2

Typical Pattern Cut by Tractor in Automatic

16

swath could be of any width from the normal cut width down
to zero. The grass edge sensing device could become
inoperative if the width of the uncut grass became zero
at some point. The microprocessor was therefore pro-
grammed to stop the tractor before the final swath was out.
In order for the lawn tractor to be capable of
executing the required pattern, several specific abilities
were required of the system. They were:

1. The tractor as modified must be capable of mowing
under manual control,

2. The microprocessor must be capable of learning the
dimensions of the rectangle to be cut,

3. The microprocessor must be capable of accurate
control of the steering wheels,

A. The microprocessor must be capable of determining
the distance traveled along a cut edge,

5. The tractor must have sensors capable of locating
the grass cut edge,

6. The microprocessor must be able to perform the
required steering algorithms to allow the tractor
to follow the grass cut edge,

7. The microprocessor must be capable of locating the
corner of the uncut lawn,

8. The microprocessor must contain a steering algorithm
capable of guiding the tractor through an accurate
270 degree turn,

9. The microprocessor must be able to determine when
the width of uncut grass is less than one meter, and

10. The microprocessor must be capable of halting forward
travel of the tractor and mower operation.

The ability to do the ten requirements stated above

were met through a combination of hardware, software and

17

mechanical systems. As there was no previous work upon
which to base decisions as to whether a specific job was
best handled in hardware or software, it was decided that
all functions that could be handled in software were
performed with software rather than hardware. This
decision allowed for maximum versatility since, in general,
software can be changed more easily than hardware. The
software was subdivided into functional blocks that were
essentially unrelated to each other in order to provide
maximum flexibility of programming. This would allow for
a major modification in one block of code without affecting
any other block. This approach made program alteration
easier but produced a less compact and possibly a program
of lesser efficiency due to similar operations being
performed in more than one block of code.

Described in the next section is the development of
the hardware and mechanical systems to implement the ten
required control or sensing operations. The following
section describes in detail the properties of the software

developed to perform the required functions.

CHAPTER V

SYSTEM DEVELOPMENT - HARDWARE AND MECHANICAL

This section describes in detail the develOpment of
the mechanical and electronic hardware systems necessary
to allow the microprocessor to control the steering of the
tractor. The development is described in a roughly
chronological fashion. However, the addition of one
circuit sometimes pointed up a weakness in another,
resulting in changes to the previously developed circuitry.
The complete development of a Specific circuit is described
as to have taken place before the next circuit is discussed
for purposes of clarity. This section will describe only
the circuit that was used in testing and major failures
that occurred during the development of the circuit. No
attempt has been made to describe all the changes in
component values or specifications that occurred in the
development of the circuit unless such changes altered
the basic concept of the circuit. Comments on the
circuits themselves and suggestions for improvements

will be discussed in a later chapter.

18

19

Linear Actuator

One critical part of an automatic guidance system is
to have the ability to turn the wheels under the control
of the five volt logic system. Several possibilities were
considered. The selected way of providing the electronic
to mechanical interface was an electric motor. This choice
was influenced greatly by the fact that the tractor was all
electric. In an agricultural application, using valves to
control the hydraulic power steering system may be more
appropriate.

The force required to start the wheels turning on
various surfaces was measured. With the tractor stationary
on dry cement, the force required to turn the wheels was
about 3,100 newtons at an eight centimeter radius from the
center of the axle. The force required on grass or with
the tractor in motion was less. To provide the required
force, a linear actuator was utilized. A linear actuator
is an electric motor driven worm gear arrangement which
provides straight line motion. A Warner linear actuator
(model DOA-10PM) rated at 4,500 newtons was obtained. It
had a stroke length of ten centimeters and an extension
rate of 2.2 centimeters per second at the maximum antici-
pated load. The.extension rate is dependent upon load.
With no load, the extension rate reached its maximum of
2.8 centimeters per second. The linear actuator was

attached to the tractor by fixing one end to the frame

20

and the other end to the vertical portion of the wheel
axle assembly. The attachment was at a point eight
centimeters from the center-of the axle. Figure 3 shows
how the linear actuator was attached to the existing
steering system.

As the tractor's manual steering system could not
be easily disassembled, the steering wheel was removed.
The conventional steering was made inoperable by the
addition of the linear actuator. Removal of the wheel
indicated to someone unfamiliar with the project that the
steering system did not function as.a conventional system.
The wheel was also removed for the safety of the operator.
The force exerted by the linear actuator caused the entire
steering system to move. If the operator reached through
the steering wheel as the system was turning, the operator
could be trapped and subjected to the full force of the
linear actuator.

In the event that the automatic control should fail,
the ability to steer the tractor should not be lost. For
this reason, the manual steering was done with a toggle
switch. This switch directly controlled the flow of
current to the linear actuator, and it was mounted in the
tractor dash. Moving the switch to the right caused the
wheels to turn, providing an increased right turn or a
decreased left turn. Moving the switch to the left caused'

an opposite reaction. When the switch was in the center

21

mnflucsoz nopmswo< hmocwq
m mmblo

 

 

/
W

e533; .8 Somme

7
a/moeaioa ~2sz

 

A30H> moev

22

position (off), the linear actuator was held in its present
condition. There was no allowance for the steering system
to return to straight automatically.

A double pole double throw switch selected either the
automatic or manual steering modes. This was necessary to
prevent the possiblilty of contradictory information from
both the manual and the automatic controls at the same
time. If the steering control system select switch were
not present, a direct short from the 12 volt supply to
ground could result.

In the automatic mode, the switching of the linear
actuator currents was accomplished by two 12 volt relays
each rated at 25 amperes. Each relay was single pole
double throw with the common connected to the motor lead.
The normally open contact was connected to 12 volts, and
the normally closed contact was connected to ground.
Energizing one relay would create a complete circuit and
caused the linear actuator motor to turn. Energizing the
other relay would cause the motor to turn in the opposite
direction. If both relays were to be erronously energized,
a complete circuit would not be made and damage could not
occur. Under such conditions, the motor would not turn.
The schematic of the linear actuator control circuit is

shown in Figure 4.

23

panopfio achusoo pogm3p0< macaw:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

w mmaon
>Ne
hopo:
hopozpo<
mmocfiq J . whom
Av - 9:99.30 scam
w\fl\o lo . . 9H. - comma
. . >NP
Io\_\h q Fl. s w 8m
1‘ mcwnooum pooaom o.
Hosea: opz<\am::mz
comzm

> NP

24

Initial Control Box

Once the linear actuator was installed and working,
it was necessary to determine the accuracy and stability
of the system. In order to do this, a control box was
built that utilized three one shots to provide three
different times for wheel movement. A one shot is an
electronic device that when triggered goes to a differnet
logic state for a.unit of time, dependent upon the value
of programming components. The control box contained nine
push buttons which were arranged in a row along the front
of the box. The center button reset all one shots to the
stable condition and thus stopped all wheel movement.
Buttons to the left of center caused the wheels to turn
towards the left, buttons to the right of center caused
the wheels to turn towards the right. The buttons on
either side of the center button caused the wheels to turn
for the output period of the shortest one shot. The next
outward pair of buttons caused.the wheels to turn for the
output period of the medium one shot. Progressing
outward, the longest one shot was next, followed by a pair
of variable buttons. The variable push buttons were wired
such that the wheels turned as long as the buttons were
pushed. The short, medium and long one shots were set to
about 0.3 seconds, 0.8 seconds and 1.5 second respectively.
The unit was installed on the tractor and information as to

the response of the tractor to various timed turns was

25

observed.

The control box was then modified to accept infor-
mation as to the position of the front wheels. Anytime
the front wheels were not centered and no button was
pushed, output voltages were produced which would cause
the wheels to return to center. Thus the linear actuator
could be activated for a unit of time, be held any desired
length of time and then would automatically return the
wheels to center. The position of the wheels was obtained
by mounting a 50 kilo ohm potentiometer to the axle such
that the case of the potentiometer was held still, while
the wiper of the potentiometer was rotated by the turning
wheels. The potentiometer position was sensed by
connecting the wiper and one end of the potentiometer to
ground, while the other end of the potentiometer was
connected through a resistor to the five volt supply. As
the wheels moved, the resistance of the lower arm of the
voltage divider changed resulting in a voltage change
between the potentiometer and the resistor. This voltage
was compared with two preset values by two voltage
comparators. If the voltage was greater than the upper
reference voltage of the upper comparator, the comparator
would have a high output that would cause the wheels to
be turned left as soon as all one shot time periods were
over and if no buttons were yet pushed. The voltage

being less than the lower reference.voltage would cause

26

a right turn under the similar conditions. The two
reference voltages were adjusted so that the wheels were
straight and the dead band was as small as possible to
provide stable operation. Satisfactory operation occurred
when the dead band was about 0.2 volts wide and centered
around 2.6 volts. The schematic of this initial controller

is shown in Figure 5.

Microprocessor Control

Once the testing of the initial controller was
finished, the next step was to put the tractor under
microprocessor control. A COSMAC VIP CDP18S711 was
selected. The COSMAC VIP is a single board video inter-
face processor. The VIP utilizes an RCA 1802 central
processing unit. The board contains two kilo bytes of RAM,
512 bytes of ROM, a keyboard, cassette tape and video
terminal interfaces, and a buffered eight bit input and
output port. A block diagram of the microprocessor is
shown in Figure 6.

The processor was to control the motion of the wheels
by creating a control word- One bit of this control word
being high would move the steering wheels to the right,
another bit would move them to the left. By generating a
series of control words, the processor could produce any
required steering algorithm- The processor also sensed
wheel position by'a method similar to that used in the

initial controller. A schematic of the wheel position

27

owpaaogom xom Hospeoo HefipneH
m mmaoHa

 

moor I unopmfimon coxhwans HH<
Foqqzm unhopwflmsmhp.woxwwszs HH<
«oovze n madcap weakness HH<

 

 

 

 

28

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

RAM , ROM , RCA
2 KBTTE<<§§§§512 BTTE<:§§§I:: 1802
4 OF TMS CDP1832 CPU
2.01.5 Q...
T” ‘ “3‘
Lg DATA BUS j
U U U F__L
OUTPUT INPUT VIDEO KEY TAPE
LATCHED LATCHED INTER- BOARD INTER-
AND FACE INTER- FACE
BUFFERED
FIGURE 6

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

RCA COSMAC VIP Simplified Block Diagram

(RCA CDP18S711)

 

29

decoding circuit is shown in Figure 7. Data describing the
wheel position was presented to the processor by means of
the buffered input port. A specific bit being high
indicated a wheel position left of center. Another bit
indicated right of center. Both bits low indicated that
the wheels were centered. A third bit went high whenever
the wheels were turned to the extreme left or right.

A program was written that caused the microprocessor
to wait for a unit of time, produce the turn right signal,
wait, stop the linear actuator, produce the left turn
signal, wait until the wheels are centered, stop the linear
actuator, and then repeat the same series.over again. All
the time delays necessary for this program were generated
by using counting loops. The count was stored in a 16 bit
register inside the microprocessor. As the maximum count
was dictated by the 16 bit register's capacity, a turn of
about 30 degrees was all that could be obtained from this
program. With this program in control of the-tractor, the
tractor went in a large.multi-sided closed figure.

The program was changed to one that performed the
same type of control, but utilized nested delay loops to
generate the delay sequences- A nested delay loop works
by using two registers for counting. Each time the first
loop was done, one of the registers was incremented. When
this register reached the desired value, the second

register was incremented and the first was reset to zero.

30

To Microprocessor

panopfio mcflcoooa coapamom H0623 pconm

 

 

 

 

 

 

 

 

 

 

 

a mmsoHa
+
pom cognac:
on< acopm
new mom 2mm on
V
so. so. go. com
com om
mam
- >m >m

>m

31

The first loop was then entered again. The nested loop was
continued until the second register was to the desired
value. This technique allowed delays of minutes or even
hours to be created. The program was adjusted so that the
tractor performed a 90 degree turn, went straight for about
three meters and then did another 90 degree turn. This
program caused the tractor to follow a pattern that was

essentially a square with rounded corners.

Distance Traveled Input

Once direct microprocessor control of the steering
function was demonstrated, the tractor was fitted with a
system to inform the processor of the distance traveled
by the tractor. This was accomplished by placing a reed
switch on the tractor frame. As the drive wheel rotated,
magnets moved past the reed switch causing it to open and
close. In theory, the output wave form from the reed
switch should consist of six pulses.per revolution. In
reality only onermagnet caused only one reed switch cycle.
The other five magnets caused two or three pulses. To
overcome this problem, a one shot was used. The time
between two adjacent distance pulses was about 1.8
seconds. The one shot was set to a period of 1.5 seconds.
The first high to low transition of the reed switch wave
form caused the one shot to start its timing cycle. It
would not accept another trigger for 1.5 seconds. By that

time, all of the spurious signals had occurred and been

32

ignored. The first pulse from the next magnet, however,
was accepted.

The output wave form from the one shot was applied to
a D-type flip-flop at the clock input. The D input was
tied high so the negative to positive transition of the
clock set the Q output high. The Q output was used to
inform the microprocessor of the fact that another distance
pulse had occurred. The Q output was tied to one bit of
the buffered input port. When the microprocessor sensed
the presence of the distance signal, it incremented the
distance value stored in memory. The processor also
delivers a high to a specific output port that resets the
flip-flop. The reset signal was then removed.

The program that resulted in the tractor following a
square pattern was modified so that rather than going
straight for so much time, the tractor would go straight
for so much distance. The changes resulted in the same
pattern, however the length of.the straight portion of the
square was less affected by small changes in ground speed.
A schematic of the distance traveled circuitry is shown

in Figure 8.

Grass Edge Sensors
Once it was proven that the tractor could be made to
follow a desired path, the next step was to fine tune the
path by providing information as to the location of the

grass edge with respect to the tractor. A grass edge

33

pfizopfio ocao>wpe mocwpmfia

m mmDon

 

 

whom
9.35:0 scum VIII!

shoe
pzmcH OBATIIII

mac Au

 

 

sopwzm ease

H mocmpmwm
a.

 

 

 

 

34

sensor that attempted to use the presence of grass to
capacitively couple a signal across the sensor proved
unsatisfactory. The capacitive reactance of grass is
quite high and dependent upon the moisture content of the
grass. The two most promising alternatives left were
mechanical or optical sensing of the grass. Optical
sensing of the grass may prove to be.more practical for

a later grass edge sensor, but the development of this
sensor was not the primary goal of this research. A
mechanical sensor was used that consisted of a microswitch
that required only about .28 newtons of force to be
activated. The selection of the microswitches as sensors
simplified the design of the.grass edge sensor because the
output of the switch was ground when the switch was closed
or pulled up to five volts when the switch was open.

The microswitches were mounted to the front of the
mower unit. A gauge wheel that adjusts the cut height of
the mower was moved inward to permit the microswitches to
be mounted on the right front corner of the mower. The
switch mountings were designed such that each switch could
be moved up, down or across a mounting bracket. This
allowed the locating of the switches with respect to one
another. In addition, the entire mounting bracket could
be moved up or down for vertical adjustment of the pair
of switches. A distance between the two switches of about

eight centimeters gave a good compromise between the amount

35

of error and the number of corrections. A smaller spacing
between the sensors would result in more precise control;
however more frequent corrections of a smaller amount per
correction would be required for stable operation. The
opposite effect occurred when the spacing was increased;
larger, less frequent corrections would be required to
obtain stable operation.

Several types and mounting orientations of the
surfaces used to activate the microswitches were tried.
The range of sensing surfaces tried varied from a small 'L'
shaped wire to a sled shaped surface of about 39 square
centimeters designed to ride on top of the grass. All the
sensor surfaces having in excess of six square centimeters
were capable of providing necessary sensing surface. A
sensor mounting position in the vertical plane was used
in all subsequent tests. This sensor was seven centimeters
long and two centimeters high. This configuration was
chosen for its ease of mounting. The sensor surface was
constructed from ten gauge aluminum. As it was not
possible to weld the aluminum plate to the microswitch
sense wire and because the plate was required to be
removable from the switch, the aluminum plate was backed
with paperboard and then mounted to the switches with tape.
This method provided adequate reliability without

permanently fixing the size of the sensor plates.

The output of the two grass edge sensing microswitches

36

represented four logic states. The desired or normal state
consisted of the inside switch sensing tall or uncut grass,
the outside switch sensing short or cut grass. If the
tractor was too far to the right, the width of grass being
cut would be less than the desired value. This condition
was indicated by both switches sensing low grass. In
order to correct this error the microprocessor caused the
tractor to turn left. If the tractor was too far to the
left, the mower would start to miss grass. This condition
was indicated by both switches sensing high grass. This
error was corrected by the microprocessor having the
tractor turn right. The final possible condition was for
the inside switch to sense low grass while the outside
switch sensed high grass. This condition could not occur
under normal operation.

If ideal conditions were used, the height of the grass
as sensed by the microswitch would be the desired logic
levels. The reality for any lawn was that the switches do
provide erroneous information. Tall grass might have been
knocked down or a low growing weed might have been present
in the tall grass, that pushed the grass away resulting in
tall grass being sensed as cut grass. Weeds that were not
trimmed at the time the grass was out can cause the sensor
to go momentarily high. In order for the system to be able
to ignor the momentary erroneous information, the outputs

of the switches were "averaged". ‘The averaging circuit

37

was required to (1) produce a digital logic level that
represented "average", (2) respond faster to an error to
normal change when compared with a normal to error change,
and (3) be stable for small changes in the "average". To
accomplish the required outputs, the circuit shown in
Figure 9 was developed. Under normal conditions, the
averaging capacitors are charged to the full five volt
supply voltage. When an error condition was sensed, the
microswitch caused the charge stored in the averaging
capacitor to be removed thus lowering the voltage across
the capacitor. This voltage level was applied to a Schmitt
trigger through a buffer. When the voltage across the
capacitor reached the lower trip level of the Schmitt
trigger, the output became low allowing the 2N4401 pull up
resistor to pull to a high logic level. One signal was
inverted before going to the microprocessor's input port.
The other signal went directly to the input port.

The return to normal followed much the same path
except the RC time constant was shorter because the
parallel diode arrangement made the effective resistance
for large differences less. As the charge on the capacitor
approached five volts, the diode was no longer forward
biased and the RC time constant was the same as discharge.
In this instance, the logic level output changed when the
voltage across the capacitor was at the Schmitt trigger's

upper trip level.

38

pfizonfio mcfimmno>< omcm mmmuo

 

b
p mmm
Aoe
0% a...
Po vzm
pa.m mus
asdaH 09
D
D
> , xmw
or
41"
AE"! mo?
Po ezm‘
x3
whom
pzch 09

 

 

o mmDUHm

oom

>m
MOP

aoom

>m

MOP

 

 

 

   

soaHSm
some“: >m
ocflmcH

MOFF

W nowwzmohowz
:NN ocfimpso

mom

 

A

sea . mm

>m

39

The Schmitt trigger was set so that the lower trip
point was at 1.02 volts. The upper trip point was set to
1.75 volts. The circuit that utilized the inside micro-
switch and sent a signal to the microprocessor that
represented a grass cut that was too small, required 2.0
seconds to go from a stable normal condition to the error
state. The same circuit would go from the error state to
the normal state in about 0.7 seconds. The other circuit
utilized the outside microswitch and sent a too large
signal. This circuit required about 0.7 seconds to go from
normal to error and about 0.3 seconds to go from error to
normal. The times selected were chosen to allow for
correction to occur as soon as required while still
responding as much as possible to an averaged input. The
inside microswitch frequently encountered grass that was
too sparse to continuously deflect the sensor plate.
Common reasons for this type of error were the density of
the grass becoming less or the grass having been knocked
down. The times required for tractor response were made
long so that the tractor would ignore, as much as possible,
the erroneous inputs. A delay in responding to a true
error condition resulted in the swath cut by the mower
being less than the desired. For the inside microswitch,
the averaging function took precedence over the response
function.

For the outside microswitch, the frequency of

40

erroneous inputs was much less than that of the inside
switch. The outside microswitch was senSing where the
grass had already been cut and was further assisted by the
fact that the tires of the tractor flatten the cut grass.
This resulted in a stable normal input and allowed errors
to be detected more quickly. This was fortunate since the
the result of failure to respond adequately to this type
of error was uncut grass being missed.

In both cases, the circuits responded more quickly to
a return to normal than to a deviation from normal. In
order for a system to be stable, the return from an error
condition to normal had to be capable of being detected and
correction stopped before the normal condition was lost.
Failure to do this resulted in the system becoming
oscillatory. The response time required for this system
to be stable was observed to be about 0.8 seconds. The
inside microswitch circuit would occasionally allow an
over correction to occur due to its 0.7 second response
time. Oscillation was stopped by the 0.3 second response

time of the outside microswitch circuit.

Drive Relay
The final interface provided for the microprocessor
to allow or prohibit travel by the tractor and mower
operation. This was accomplished by placing a normally
open relay in series with the existing safety interlock

system of the tractor. The microprocessor completed the

41

safety interlock system by presenting a 1 to a specific bit
of the output port. The presence of the 1 would cause the
normally open relay to close. If all other interlocks were
in the required state, the circuit was completed and power
was supplied to the tractor system. The schematic of

this circuit is shown in Figure 10.

With the completion of the drive relay circuit, the
microprocessor was connected with all the interface
circuits required to control the tractor. In addition to
the previously described inputs and outputs, several logic
level signals were provided to the microprocessor to
complete the interface with the outside world. One bit of
the input word was allotted to a logic level input that
represented future obstacle inputs. This bit was normally
low. When it went high, all machine operation was halted
under regular program control. A second input bit
informed the microprocessor of the operator's desire to
have the microprocessor "learn" the size of the lawn. In
addition, one of the microprocessor's sense lines (EFA)
was utilized to inform the microprocessor if the first
or the second side was being learned. This same sense
line was used to signal the microprocessor to return from
a machine generated shutdown, thus allowing normal
operation to resume. A final input to the microprocessor
was via the interrupt line. When this line was pulled

true, the microprocessor immediately jumped to the

42

Hfioo
awamm nosom
nopowpe oe

 

 

e.

>©m

pnzocflo madam
o. mmeoHe

b. .D

«I
Fosezm .N\‘ so?

w

seams swam
<m .>NP

cough >NP

o>Hnn

voo zr

voo.zP

\
/

whom
unease
scum

ccpfizm
auspom
scum

43

shutdown routine regardless of normal program operation.
Table 1 shows the bit assignments of the input port and

Table 2 shows the bit assignments of the output port.

Operator Controlled Inputs

Six switches were used by the tractor operator to
interface with the automatic steering system. The first
switch was the main power on-off switch. This switch was
turned on at the start of operation and remained on until
operation was complete. The second operator used switch
selected manual or microprocessor control of the linear
actuator. This switch was mounted in the box that also
contained the linear actuator drive relays. This switch
was in the manual position whenever the operator desired
direct control of the steering. In the auto position, the
microprocessor had full control of the steering system.
The switch had a center off position that disabled both
manual and automatic steering. This switch was a double
pole double throw toggle with three stable states.

When the steering select switch was set to manual,
the operator controlled the linear actuator by the use of
a toggle switch mounted in the dash. Moving the toggle to
the left would cause the tractor to turn the wheels towards
the left. Similarily, moving the switch to the right
caused motion of the wheels towards the right. In the
center, the wheels were held in their current position.

The switch was a double pole double throw toggle switch

Bit Number

44

TABLE 1

INPUT PORT ASSIGNMENTS

Low Logic Level

No obstacle present

Wheels not
turned maximum

Wheels not right
of center

Wheels not left
of center

No new distance
pulse

Learn mode

Outside grass
edge sensor normal

Inside grass edge
sensor normal

High Logic Level

Obstacle present

Wheels
turned maximum

Wheels right
of center

Wheels left
of center

Distance pulse
received

Not learn mode

Outside grass
edge sensor abnormal

Inside grass edge
sensor abnormal

Bit Number

45

TABLE 2

OUTPUT PORT ASSIGNMENTS

 

 

Low Logic Level High Logic Level
Open safety Complete safety
interlock circuit interlock circuit
No turning left Turning left
No turning right Turning right
Distance input normal Reset distance
flip-flop
Operation normal Shutdown exists

Not used in present system
Not used in present system

Not used in present system

46

with one stable state and two momentary states.

The fourth switch was a four pole six position rotary
switch. This switch was used to inform the microprocessor
of the start and end of sides to be learned and to initiate
automatic operation. The connections to the rotary switch
are shown in Figure 11. The operator drove the tractor to
the lawn in position one. While still in position one, the
operator cut the first two sides of the rectangle and
positioned the tractor at the start of the third side. The
microprocessor was then informed of the fact that it should
"learn" the length of this side when the operator placed
the rotary switch into position two. As the third side was
being cut, the microprocessor learned the length of that
side by counting distance pulses. Just before the end of
the third side, the microprocessor was told of the
completion of that side by moving the rotary switch to
position three. The last side of the rectangle was
"learned" by placing the rotary switch into position four
at the start of the side and to position five at the
completion of that side. After the tractor had been
positioned back at the start of the first side, the
steering selector switch was placed into the auto mode and
the rotary switch was placed into position six. The
tractor was now under microprocessor control. Upon
completion of the lawn, the rotary switch was returned to

position one and the steering select switch returned to

47

capmamnom nowfizm mumpom

PF mmaon

 

 

 

\ eel].
.39; E om. / o/
. om
o—om
Q0
on
Achmoqv AM aw i
atom .39: as [a
olllL
O— N
00
call
peachflo mmamm.h, ow
03.5 3. / Lt/cmll
on

filllL
E: molIIL

 

nommooonQOAOHz \\ am
on

mo pomom\csm ca // AY/ll

v

 

 

48

manual so that the tractor could be transported to storage
or to another lawn.

The final two switches available to the operator were
both momentary push buttons. One was mounted atop the
microprocessor box and caused an immediate machine shut-
down. The second was mounted near the rotary switch and
caused the microprocessor to leave shutdown and return to

normal operation.

CHAPTER VI

SYSTEM DEVELOPMENT - SOFTWARE

The concept in the development of the software was
to keep sections of program independent. This was
accomplished by making use of five registers as program
counters for various blocks of program. The program moved
from one block to another by changing the program counter
to a different register. In the RCA 1802 microprocessor,
a total of sixteen registers were available. The software
utilized ten of the sixteen registers. The first five
registers were program counters for the various blocks of
program. Register two was called the X register and its
function was to point to various locations in memory. An
output function, for instance, sent the value of a memory
location pointed to by the X register. Five other
registers were used to store data and serve as a counter
for delay loops. The register allocation is shown in
Table 3.

The VIP system used in this project stored the entire
program in RAM. The system contained two kilo bytes of
memory. When operation of the program began, the first
location read was location 0000 (hex). For that reason,

the first portion of the prOgram to be executed must start

49

R(0)
R(1)
R(2)
R(3)
RU.)
R(5)
R(6)
R(7)
R(8)
R(9)
R(A)
R(B)
R(C)
R(D)
R(E)
R(F)

50

TABLE 3

REGISTER ALLOCATION

Main and initial program counter
Shutdown program counter

X Register, pointer

Steering function program counter
Learn program counter

Not used

Not used

Not used

Not used

Not used

Not used

Output word

Distance 1 0:0

Distance 2 Q=1

Outer loop counter

Inner loop counter

51

at 0000 (hex). The location of the remainder of the
program inside the available memory was arbitrary. Hex
locations 03F9, 03FA and 03FB contained a copy of the
contents of three registers internal to the microprocessor.
This was necessary because the internal register contents
could not be displayed. In order to verify the contents of
these registers, their contents had to be copied to RAM and
then the contents of the RAM location displayed. These
three locations served no purpose other than an aid in
evaluating system operation. In all cases, the amount of
memory space allocated to the program block was more than
the amount that was actually used. The allocation of
memory to various blocks of the program is shown in Table
4. This figure also contains the actual number of memory
locations used by the program.

Discussion of the software itself can be conducted in
two ways. In one way, the actual mnemonics of the specific
microprocessor can be referenced. This type of discussion
tends to be very detail oriented and required a fairly good
understanding of the microprocessor and its language. The
other basic way to discuss does not utilize the mnemonics
but instead uses a statement to represent the function of
a small group of individual mnemonics. In most cases, the
use of descriptive statements allows the logic and flow of
the program to be more clearly presented. This discussion

will not use mnemonics to describe the software.

Location
(Hex)

0000-003F
0040-00FF
0100-02FF
0300-0389
0390-O3BF
03C0-03EF
03F0-03F8
03F9
03FA
03FB
03FC
03FD
03FE
03FF

52

TABLE 4

MEMORY ALLOCATION

Assignment

Initialize program
Main program loop

Turn programs

Open

Learn

Shutdown

Data storage open
Contents of B register
Contents of O register
Contents of D register
Required distance side
Required distance side
Output word

Input word

Number of
Locations Used

58
54
278

47
11

53

The program was divided into eight main blocks shown
in Figure 12. The program started at memory location 0000
(hex) with the initialization program. This program
performed the bookkeeping functions necessary to the proper
operation of the remaining blocks of program. If the
microprocessor sensed that bit 5 of the input word was low,
the program counter was changed to register four. This
caused the learn program to be executed. In the event that
bit 5 was to go high after the learn program had been
entered, the program branched to shutdown.

In the event that the initialization program did find
that bit 5 of the input word was high, the program executed
next was the main loop program. This program read a data
word from the input port and examined certain bits of the
word. If bit 0 was high, the program branched to shutdown.
If bits 6 or 7 were high, the program branched to the turn
selection program. If bit 4 was high, the distance count
was incremented by one and the acknowledge bit was made
high for a short amount of time. If none of the bits
mentioned were high, the program returned to get a new data
word and the whole process started over again.

The shutdown program was entered by changing the
program counter to register one. The shutdown routine
removed the output that kept the safety interlock system
closed, thus stopping the tractor. The program then waited

for the sense line EF4 to become true. When that happened,

 

 

 

 

 

 

 

54

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

C START D
INITIALIZATION <
PROGRAM
LEARN
3’ PROGRAM
.MAIN LOOP ‘
' 9 PROGRAM
‘ 7E I SHUTDOWN
' - ' ' >1 PROGRAM
AL
TURN SELECTION
PROGRAM
CORRECTION RIGHT
PROGRAM
CORRECTION LEFT
‘5’ PROGRAM
CORNER TURN
IE’ PROGRAM
FIGURE 12

System Floijhart

 

 

 

 

 

55

the program continued with the initialization routine.

The turn selection routine was entered by changing the
program counter to register three. The turn selection
looked at the type of error that had been sensed by the
grass edge sensor. It then selected one of three steering
algorithms based upon the type of error and the distance
remaining to be cut. The appropriate algorithm was
executed and control returned to the main loop program. A
more detailed flow chart of the initialization program is
shown in Figure 13. When the central processing unit was
first turned on, the contents of the various registers were
unknown. In order to assure proper operation of the
program, the registers must be loaded with the correct
values. This was the bulk of the work of the initial-
ization program. This program also put out the signal
required to close the tractor's safety interlock system.
Finally, the program branched to either the main loop
program or the learn program, depending upon the condition
of bit 5 of the input word.

Figure 14 shows a more detailed flow chart of the main
loop program. The program read a data word. If bit 0 was
high, the program counter was changed to register one
causing the shutdown program to be executed. If bit 6 or
7 was high, the grass edge sensors had detected an error.
The main loop program responded by changing the program

counter to register three where the error was evaluated and

 

C START?
I

 

 

R3 = 0101

 

TURN PROGRAM

 

 

 

R1 = 03DF
SHUTDOWN PROGRAM

 

J7

 

 

R2 = 03FC
X REGISTER

 

AL

 

 

ZERO RC AND RD

 

 

 

 

 

 

OUTPUT INITIAL
WORD AND STORE

L. I .

 

ENTER FROM
SHUTDOWN

 

 

 

I

 

 

R4 = 0399
LEARN PROGRAM

 

 

   
 

 
  

IS BIT 5
HI-LEARN

 

 

BRANCH TO
LEARN PROGRAM

 

 

 

BRANCH TO
MAIN LOOP

 

 

FIGURE 13
Initialization
Flow Chart

 

 

57

 

C START)
‘ NV

 

 

 

INPUT DATA WORD <
TO 03FF

 

 

 

IS BIT 0 H YES BRANCH TO
SHUTDOWN SHUTDOWN PROGRAM

 

 

 

NO

 

YES BRANCH TO TURN
SELECT PROGRAM

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

IS BIT 4 HI YES INCREMENT
DISTANCE SELECTED DISTANCE
NO OUTPUT

ACKNOWLEDGE
/
FIGURE-14

Main Loop Flow Chart

58

corrective steering functions executed. Bit 4 high
indicated that another distance sensor had been closed,
setting the distance flip-flop. The main loop responded
to this occurrence by incrementing either register 0 or
register D. Register 0 was incremented if the internal
flip-flop of the processor was reset, otherwise register D
was incremented. Following this, the microprocessor sent
a high to bit 4 of the output port. This reset the
distance flip-flop. Bit 4 was then returned to its normal
low position. Failure to find any of the bits checked to
be high, resulted in the main loop program returning to
its start to read a new data word.

A detailed flow chart of the turn selection routine is
shown in Figure 15. This routine was entered whenever any
error had been detected by the grass edge sensors. If bit
6 of the input was high, the error was that the mower was
attempting to out too much grass. In response to this
error, the correction right program was executed. If bit
6 was not high, the turn selection program compared the
distance traveled on the current side to the length of that
side. If the distance traveled was not greater than or
equal to the length of the side, the turn selection program
branched to the correction left program. If the tractor
had traveled the length of the side, the edge was assumed
to have been lost due to the tractor reaching the corner.

The turn selection program responded by starting the corner

59

 

C START j

 

BRANCH TO
CORRECTION RIGHT

 

 

 

 

BRANCH TO
CORNER TURN

IS SELECTED

 

 

 

 

 

 

BRANCH TO
CORRECTION LEFT

 

 

FIGURE 15

Turn Selection Flow Chart

60

turn program.

The program, as described, allowed the microprocessor
to choose one of three different steering functions. Two
were essentially mirror images of each other. They were
the correction right and the correction left programs.

The flow charts for both programs are shown in Figure 16.
For both turns, the microprocessor produced a signal to
start the wheels turning in the appropriate direction.

The wheels were allowed to turn for an amount of time
controlled by the program. The wheels were then held in
that position for another microprocessor controlled period
and finally the wheels were returned to straight forward.
The periods actually used were about 0.5 seconds, 0.7
seconds and 0.5 seconds, respectively. The entire process
must be completed before 1.8 seconds had elapsed. This

was because at the given ground speed of the tractor, a
distance pulse would occur every 1.8 seconds. If the turn
sequence was allowed to take more than 1.8 seconds, the
first distance pulse would set the distance flip-flop, the
next distance pulse would have no effect upon the set flip-
flop and thus be lost. A system that gave good control but
failed due to this 1.8 second requirement, was one that
turned the wheels for a unit of time and then left them
there until the grass edge sensors reported normal
conditions. Under ideal grass conditions, this type of

control was acceptable. However, distance pulses were lost

 

C START?
\E

 

 

OUTPUT RIGHT TURN

FOR UNIT TIME

 

LSTART )

 

 

OUTPUT LEFT TURN
FOR UNIT TIME

 

 

l

Al/

 

 

 

DELAY FOR.
UNIT TIME

 

 

DELAY FOR
UNIT TIME

 

 

I

 

 

 

 

OUTPUT LEFT TURN
UNTIL.CENTER

 

 

I

 

OUTPUT RIGHT TURN
UNTIL CENTER

 

l

 

 

 

BRANCH TO
MAIN LOOP PROGRAM

 

 

 

Correction Right
Flow Chart

BRANCH TO
MAIN LOOP PROGRAM

 

FIGURE 16

Correction Left
Flow Chart

 

 

 

62

under more common grass conditions resulting in inaccurate
distance traveled counts. The main problem with an
inaccurate distance traveled count was that the corner of
the uncut lawn was no longer where the microprocessor
calculated that it should be. Each pass through the
correction program resulted in a correction of about three
centimeters. Errors that required more than that amount of
correction resulted in the wheels appearing to turn out,
stop, return to center and then immediately turn back out.
In reality, the microprocessor had updated the distance
input, read a new data word, checked all other input bits
necessary and then the main loop program returned control
to the correction program.

The corner turn was one of the more critical portions
of the software. The turning of the corner required about
one minute. In order for the turn to be considered
successful, the tractor had to be brought back to the uncut
grass edge with enough accuracy to permit the grass cut
edge to be located by the sensors within a reasonable
amount of time. The flow chart of the program that gave
acceptable results is shown in Figure 17. The program
called for the tractor to travel with the wheels held in
the straight position for a period of time controlled by
the microprocessor. The tractor traveled about 2.1 meters
in this time. The wheels were then turned for nearly a

maximum right turn. The wheels were not turned full

63

 

C START j
- I

GO STRAIGHT FOR
UNIT TIME

AD

OUTPUT RIGHT TURN
UNTIL LIMIT

\I/

TURN FOR
UNIT DISTANCE

T

OUTPUT LEFT TURN
UNTIL CENTER

\l/

DECREASE VALUE OF
SIDE BY 3 UNITS

A

POINT TO
OTHER SIDE

A

RETURN TO
MAIN LOOP

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

FIGURE 17

Corner Turn Flow Chart

64

maximum for fear that a slight shift in the sensing system
could result in full linear actuator force being applied
to an immobile system. Damage to both the linear actuator
and the steering system could result.

Once the wheels were fully turned, the tractor
followed the circular path for 24 distance pulses. Sensing
the distance traveled pulses from the outside turning wheel
gave the largest count possible. The measurement was
initially attempted upon a time basis. The time-based
method proved to be very sensitive to ground speed varia-
tions caused by changes in battery voltage or differences
in the force to move the lawn mower through varying lawn
conditions. For instance, a turn going up a slight hill
will be longer in time than one on level ground. Once the
tractor had been repositioned near the corner of the uncut
grass, the microprocessor performed operations that
decreased the value of the side by three distance units and
and changed the flip-flop internal to the microprocessor so
that the other side length would be used next. Control
then returned to the main loop program that caused the
tractor to locate and follow the grass cut edge using the
two correction programs. When the grass was wet, it was
possible for the slip of the drive wheel to be such that
substantial errors were created. At the same time, the
steering wheels could not provide adequate traction to

provide the desired amount of turn for each unit of travel.

65

The combined effects of the two errors resulted in an
unacceptable turn on wet grass.

The shutdown program flow chart is shown in Figure 18.
This program was reached by changing the program counter to
be register one. This could occur under program control or
by application of an external interrupt signal. The first
event that happened was the output signal that kept the
safety interlock circuit closed was removed. This caused
both tractor and mower operation to cease. The micro-
processor then waited at one program location until a reset
signal was applied to one of the sense lines called EF4.
Upon receipt of this signal, the program control was
transferred to the initialization program that put out a
new control word to allow machine operation, checked for
learn mode, and if learn mode was not required, transferred
control to the main loop program.

The final block was the program that learned the size
of the lawn. This program was entered by changing the
program counter to register four. The flow chart of the
program is shown in Figure 19. This program started by
reading a new data word and checking to see if a distance
pulse had occurred. If no data pulse was present, the
program read a new data word. When a pulse was present,
the microprocessor incremented the appropriate count. The
microprocessor selected the count to increment based upon

the status of the EF4 line. When EF4 was high, side one

66

 

C STARTQ
\l/

 

 

'OUTPUT SHUTDOWN
WORD

 

 

 

 

 

IS EF4 NO
TRUE

YES

 

 

BRANCH TO
INITIALIZATION

 

 

FIGURE 18

Shutdown Flow Chart

67

 

C START )

 

 

      

 

IS BIT 4 HI
DISTANCE

 

 

 

 

 

INCREMENT SIDE

 

 

INCREMENT SIDE

 

 

 

 

 

 

 

 

 

 

ONE COUNT ‘5’ TWO COUNT
lJ
OUTPUT
ACKNOWLEDGE
IS BIT 5 H BRANCH TO

     

LEARN

 

SHUTDOWN PROGRAM

 

 

 

FIGURE 19

Learn Fl

ow Chart

 

 

 

68

was used. Grounding EF4 would cause side two to be used.
The distance flip-flop was then reset by the acknowledge
signal; the microprocessor then verified that the learn
mode was still selected. In order to prevent learn mode
problems from going unnoticed, the program caused the
shutdown routine to occur whenever the learn mode signal
was removed. The normal operation of this program was
halted by moving the rotary switch. This halted all
microprocessor activity and returned full manual control.
The microprocessor considered the most recently updated
value to be the length. When the microprocessor activity
was halted, the most recent value was the one retained.
The retained value was used to determine the location of
the corner in automatic mowing. Because the tractor was
driven farther on the programming cut than was the actual
length of the side that was to be cut, the value stored
must be adjusted. The method selected to do the adjust-
ment was to start the count as a value of negative six.
The figure of negative six was selected by combining the
possible errors. An error of three resulted from the side
being shortened by one swath width. As the distance flip-
flop was set due to the tractor travel during the turn,
one distance unit was removed to compensate for this
initial error.

Finally, both the initial learn cut and the actual

automatic cut would not usually be even distance increments

69

which resulted in an error. For instance, if the actual
length of the side was 10.4 units, the tractor could
measure this side to be either 10 or 11 units, depending
upon where the first distance pulse occurred in relation
to the start of the edge. If the distance pulse occurred
after 0.1 units of travel, a length of 10.4 units would
involve 11 pulses. They would occur at 0.1, 1.1, 2.1, ....
9.1 and 10.1. If on the other hand, the first distance
pulse did not occur until after 0.8 units of travel, a
total of 10 pulses would occur at 0.8, 1.8, 2.8, .... 8.8
and 9.8. The eleventh distance pulse at 10.8 was never
reached. In order to insure that the measured length was
not greater that the actual length, one was deducted from
the count. This type of error resulted in the tractor
possibly entering the turn too soon. In reality, the
problem was not as severe as might be expected because the
actual turn program did not start until both the distance
had been traveled and the grass edge sensors both sensed
low grass. In the event that the grass edge sensors
determined what was actually an error in following the
grass edge after the distance had been traveled but before
the actual corner had arrived, the tractor would enter the
corner turn program too soon. In most cases a triangular
shaped piece of lawn was missed. If the 270 degree turn
was started 0.3 meters too soon, an area of grass would be

left uncut that would total about 0.3 square meters. A

7O

substantial portion of the grass missed in the previous
turn was cut during the next 270 degree turn.

The entire software package was composed of various
blocks of code that were treated by the main loop program
as subroutines. This enabled the development of software
that was easily understood, but that tended to have lesser
microprocessor efficiency and control system efficiency.
The software as described was capable of controlling the

tractor to the desired specifications.

CHAPTER VIII

RESULTS AND DISCUSSION

The goal of this project was to produce a lawn tractor
that was capable of (1) learning the size of the lawn, (2)
cutting the lawn, (3) shutting itself off, and (4) of
operation in other than the automatic mode. The tractor
as described was able to meet the goal. There is much room
though, for improvement. The tractor was not tested on
nearly as many rectangular lawns as would have been
preferred. The main reason for the lack of testing was
cold wet weather that inhibited the growth of grass. The
tractor has been operated enough to allow conclusions as to
the ability of the system to mow an idealized rectangular
lawn. In total, twenty rectangles of lawn were mowed. Few
of the tests were performed to evaluate the total perfor-
mance, most of the tests were conducted to evaluate
specific portions of the tractor's operation. For
instance, some test areas were only three swath widths wide
and were used to observe the shutdown function. The
ability of the tractor to do some specific task was tested
by situations other than normal operation. Two tests were
run to evaluate the ability of the tractor to follow the

grass cut edge. In one test, a grass edge of about 30

71

72

meters was established. The tractor was positioned on the
edge and allowed to follow for the total length. The
tractor was then manually returned to the original end and
allowed to follow the newly created edge. In the first
series of runs, the tractor followed a straight path.

After five passes, the edge was no longer exactly straight.
However, the actual path was never more than 30 centimeters
from the ideal path. The edge contained two major varia-
tions from the desired straight line. In both cases these
errors resulted from a correction to the left that was the
result of errors by the grass edge sensors to correctly
determine the location of the grass cut edge. One error
was caused by a bare spot, the other by an initial paper
sensing plate failure. When these two problems were
removed, the final cut edge varied no more than ten centi-
meters from the ideal. 0f the attempted grass area, 100
percent was out.

A second edge was created that included a jog to the
right of about 30 centimeters. This jog occurred in about
two meters of travel. After eight passes, the jog was
still present, but had "moved" in the direction of tractor
travel. During this test, the lawn tractor cut in excess
of 99 percent of the attempted lawn. One narrow sliver of
lawn was left uncut when a bare spot caused the tractor to
correct left twice. In reality, no correction should have

occurred. The tractor was unable to return to the correct

73

grass cut edge in time to avoid missing grass. The missed
spot was about 250 square centimeters.

A second test was performed using a similar method
except that this time the total length of cut was about 15
meters. For this test, the deviation from the desired path
was measured at 61 centimeter intervals. The initial cut
edge varied no more than three centimeters from straight.
The total of individual errors divided by the number of
measurements gave an average error of less than two centi-
meters for the first automatic cut. The maximum deviation
was eight centimeters. The second pass had an average
error of two centimeters and a maximum deviation of ten
centimeters. The third through seventh passes had average
errors of 0.3, -0.3, 2.0, -0.4 and 2.0 centimeters respec-
tively. The maximum deviation was 10, 13, 13, 15 and 18
centimeters respectively. The maximum deviation that
resulted tended to occur in about the same point in the
distance from the origin. For instance, the maximum
negative deviation occurred at both 7.9 and 8.5 meters from
the origin for the first pass. The maximum negative
deviation for the seventh pass occurred at 9.7 and 10.3
meters from the origin. In general, the edge shape tended
to "move" along the direction of travel. The mower cut
99.95 percent of the total grass attempted. When the mower
missed grass, it was due to a response to invalid grass

edge error signals. A grass edge sensor that would

74

differentiate between cut grass and no grass would provide
a means to greatly reduce this type of error.

Other tests performed on only parts of the system
included Operating the tractor over a fixed course while
in the learn mode. As the dimensions of the course were
known, the count that should be stored in the two length
registers could be predicted upon completion of the course.
The actual value held in memory was compared with the
predicted value. The actual value was equal to, or one
less than the predicted value in all but one case. In
that one case, the actual value was one greater than the
predicted value. The plus or minus one error is a result
of the method used to measure distance traveled. A
distance pulse was generated by the circuitry for each unit
of travel. The unit of measure for the tractor was very
close to 0.3 meters. The distance pulse could occur at any
time during the first unit of travel, and was dependent
upon the location of the magnet with respect to the reed
switch at the start. If the magnet was close to the reed
switch, the distance pulse would come at the beginning of
the unit of travel. Thus, for each full unit of travel,
one distance pulse would occur. For a partial unit of
travel, one distance pulse may or may not occur depending
upon the positioning of the pulse within the travel time.
Frequently, the distance measured count was one less than

expected due to the timing errors just described. The one

75

measurement that was too large was attributed to the
tractor being driven just slightly longer than the desired
length. The distance pulse must have occurred very soon
after the start of travel, allowing the final distance
pulse to occur in the slight overshoot.

Another way in which the tractor was operated without
actually being in full automatic control was to manually
control the grass edge sensors such that an idealized grass
edge was simulated. This method was used to test the
ability of the system to start and execute the 270 degree
turn. With the portion of the program that caused the side
length to be decremented three times disabled, the tractor
would follow a fixed rectangular pattern. With the program
modified, the ability of the tractor to do multiple
accurate 270 degree turns was evaluated. A square of about
three meters on each side was outlined on a lawn with
string. The grass edge sensors were manually operated to
provide the signals that would have occurred if an ideal
grass edge was on the string. The size of the square was
selected to represent conditions that would occur when the
lawn was approaching completion of mowing. After one
complete revolution only about 1.5 square meters remained.
As the length of the lawn became smaller, the error created
by the 270 degree turn not being exact.became more
apparent. This was because less time was available to

allow the tractor to orient itself correctly upon the grass

76

cut edge. If the tractor had not fully corrected the error
resulting from one turn when another turn begins, the
second turn would have the sum of the initial error and the
turn error. The amount of distance traveled to overcome a
turn error was directly related to the magnitude of the
error. A turn that resulted in 15 centimeters of offset
and a few degrees of misalignment would usually be
corrected in about two meters of travel.

The three meter square was thus a severe, but not the
worse possible test of the 270 degree turn. With the grass
edge signals being controlled, the tractor went three times
around the square for a total of twelve turns. After the
twelve turns, the tractor was still on course and capable
of continued travel. The test was stopped because only six
turns would be required to finish this size lawn. At the
time of this test, the 270 degree turn program was
conducted entirely on a timed basis. As shown, this worked
well under some conditions. However, the quality of the
turn was very dependent upon the speed stability of the
tractor. Accuracy similar to that just described was not
consistantly possible. For that reason, the turn program
was changed to where the turn was executed using a fixed
number of distance units. This change resulted in more
consistant but less accurate turns. This was a result of
the distance measuring problems that were previously

described. The turn ended when the final distance pulse

77

occurred. As the distance pulse occurred anywhere during
the unit of travel, the turn could be finished at any
point. The total number of distance pulses required for
the turn was 25. The turn could thus end anywhere between
24 and 25 actual units traveled. This error appeared
acceptable when viewed in the total accuracy of the system.
The test just described was not repeated for the revised
270 degree turn.

Only six rectangular lawns were attempted under full
operating conditions. 0f the six, only one was fully out
without a substantial error occurring. In one case, the
outside grass edge sensor plate shifted and would not
allow the microswitch to return to the normal condition.
Another attempt was hampered by the misadjustment of the
grass edge microswitches with respect to the cut height.
The sensors were set low enough that cut grass was deter-
mined to be uncut grass when the tractor went through a
slight dip. This error resulted in three undesired right
corrections occurring. In both cases just described the
tractor was stopped, corrections made and the tractor
restarted from the same point. In one attempt, the size of
the lawn was such that the third turn took the tractor in
such a way that the front left hand corner struck a sign
post located near the test sight. The tractor was stopped
in such a way that the tractor could not simply be freed

and returned to automatic operation. The size of the lawn

78

had to be relearned before the return to automatic could
occur. The tractor finished the lawn.

Another attempt was marred when a piece of cut grass
landed and remained across the potentiometer leads. This
resulted in the wheels not being straight when indicated by
the potentiometer and the 270 degree turn not being as
sharp as desired. The tractor returned to normal operation
when the grass was removed. Prior to the location of the
problem, three 270 degree turns were executed. As the
distance between the second and third turn was not long
enough to allow the effects of the second turn error to
be eliminated before the third turn began, the third turn
resulted in a large positioning error. The tractor was
stopped, and repositioned manually, as well as the problem
located and corrected before automatic operation resumed.
The tractor finished the lawn.

In the final attempt, the problem resulted from two
270 degree turns separated by about two meters both
starting the turn too soon. This occurred when the tractor
was not stabilized upon the grass edge. The turn began
when the tractor had traveled the required distance and the
grass edge sensors both sensed cut grass. Due to the
distance measurement problem previously discussed, in order
to insure that the distance held in memory would be reached
before the actual corner was reached, the memory distance

was less than the actual distance. Under certain

79

conditions, the memory distance could be reached almost 61
centimeters before the corner. When the tractor was not on
the grass edge, the turn could start almost 61 centimeters
early. Two of these errors in rapid succession created the
problem. The tractor was stopped, manually positioned on
the grass edge and automatic Operation resumed. The
tractor finished the lawn.

For all the six cases, the goal of actually mowing in
excess of 95 percent of the attempted amount was obtained.
In spite of the problems just described, the tractor was
able to prove that the concept used in the development of
this tractor was valid. From the beginning, it was felt
that the microswitch grass edge sensors were not the best
choice. They were, however, the simplest choice. If the
problems to which the microswitch grass edge sensors
contributed were eliminated, two more complete lawns would
have resulted. Physical problems unrelated to the concept
of the controller caused two problems. Neither problem
should occur again. The final problem, while not caused
directly by the grass edge sensors, would be avoided by a
grass edge sensor design that looked over a broader area
for the grass edge and directed the controller to perform
one of several correction algorithms that would eliminate
the error based upon its size. With an improved grass edge
sensor and the knowledge gained from the first six tests,

at least five and probably all six rectangles could have

80

been mowed without any more than the minimum required

assistance from the operator.

Grass Edge_Sensor

The grass edge sensors as described and utilized would
never be suitable for any more than the demonstration of
the Concept. What the microswitches actually responded to
was the ability of the grass to push the sensor plates
back. The correlation between the ability to push the
sensor plate back and the height of the grass was not all
that good. Uncut grass that had a low number of blades per
unit area might not have the strength required to overcome
the resistance of the microswitch. Clover or other plants
growing in the lawn could force the grass out while not
themselves growing tall enough to operate the microswitch.
The microprocessor did not need to know if the grass can
push back, it needed to know if the location was cut or
uncut. A non-contact form of sensor would best achieve
this requirement. Systems that have been discussed as
possible solutions to this problem included several optical
systems and an ultrasonic system. The optical systems
ranged from a simple light beam sensor where the grass
breaks the beam to a digital camera whose output is inter-
rupted by the microprocessor. An advantage of optical
systems over the microswitch was that there were no
moving parts. The ability of the microswitch to withstand

'years of use was questionable.

81

Another problem with the grass edge sensors as
described was the ability to provide useful information.
The microswitches could only provide information as to the
direction of the error. With only this information avail-
able, just one COrrection in each direction could be
utilized. The correction had to be small enough to provide
stable operation when the error was the smallest detectable
error. Thus when the tractor was faced with a step change
of fairly large proportions, the tractor responded with
multiple small corrections. The addition of just two more
sensors would allow the microprocessor to determine if the
error was large or small. The microprocessor could respond
to a large error with a much larger relatively crude
correction. This could be followed if necessary with a
small correction to fine turn the following of the grass
edge. Once the grass edge was located and the tractor on
course, only the small corrections would be required to
follow the edge. The tractor would be able to respond
quicker to a large error of the type that might occur when
a 270 degree turn was executed, for example.

Another problem with the grass edge sensors as
utilized in this development was that no information as to
heading errors was available. A different correction was
required to correct an error that had the tractor going

parallel to the desired path when compared to an askew

heading. A heading correction required only that a turn in

82

one direction be executed and then the wheels returned to
straight. The offset error required that the heading of
the tractor be changed twice. The first eliminated the
offset, then the second restored the tractor to its
original heading. Information as to heading errors could
be obtained by having two pairs of sensors located such
that the first pair was in front of the second pair.
Offset errors would be indicated by both pairs of sensors
showing the same error condition. If only one of the two
pairs showed an error, the error would be a heading error.
Both types of errors could be corrected by heading correc-
tions. However, a correction designed for offset errors
could correct the problem with one correction, whereas at
least two heading error corrections were required. A
logical way to determine offset and heading errors would
be to utilize a four by two matrix of sensing devices.

The final problem with the grass edge sensors was
that the sensors could not differentiate between cut grass
and bare spots. If a bare spot was encountered in the
lawn, the tractor should ignore the normal tendency to do
a correction. The tractor should continue straight if it
was on the grass cut edge when all grass was lost. The
tractor should stop if no grass is found after some fixed
amount of travel had occurred. This information could be
gained by placing an optical sensor below the normal cut

height of the lawn mower. This sensor would normally be

83

blocked by the grass, however the beam would be completed
when a bare spot was present. Another possible approach
would be to use a special digital camera to look for the
grass edge. This technique appears promising, but will
require considerable work to deveIOp. The use of an

ultrasonic sensor does not appear to be very promising at

this time.

Distance Traveled Circuitry

Magnets and a reed switch of the type used in a home
burglar alarm were used to provide the pulse train that
indicated travel. This arrangement resulted in one pulse
for every 0.3 meters of travel. The problem with this
arrangement was that the best accuracy that could be
obtained was plus or minus one distance unit. As this
error occurred in both the learn measurement and the length
of travel measurement, the total error could be as much as
two distance pulses (about 0.6 meters). To compensate for
this problem, the distance at which the tractor could start
its turn could be as much as 0.6 meters from the corner.
One way to limit the effect of this problem would have been
to increase the number of pulses that occurred for a unit
of travel. If the number of distance pulses were increased
by a factor of ten, the maximum error would be reduced to
six centimeters. This was not possible with the current
concept for two reasons. First, it was not possible to

place nine magnets between the existing ones and have them

84

all work the same reed switch. Secondly, the micro-
processor was required to respond to each distance pulse
which would now occur every 0.18 seconds. This, in itself,
was no problem but making all other routines less than 0.18
seconds would be a nearly impossible task. No worthwhile
correction of a steering error could be accomplished in
0.18 seconds. The first problem can be solved by doing the
sensing of distance before the transmission. The input
signal would thus be multiplied by the reduction of the
transmission. A different type of sensor may be required
-to work at the new speeds. The second problem required a
change in concept to solve. As the number of pulses was
increased to improve accuracy, the use of a system to
divide the number of pulses down was unproductive. The
only logical way was to allow an external circuit to keep
track of the number of pulses and report that number to the
microprocessor upon command. This arrangement would allow
the accuracy of the higher number of pulses while not

using any more microprocessor time than was required by

the old system.

Wheel Position Sensor
This system worked and provided the required amount of
accuracy. A two wire shielded cable was utilized in this
situation. A better alternative might have been to use a

three wire cable. This would allow a separate wire to be

85

used to return the wiper voltage to the control unit. This
change would have a couple of advantages. The system as
previously described was not a linear system. The two
extreme conditions produced output voltages of 0.2 and

3.2 volts. The straight forward position produced 2.6
volts. The amount of wheel rotation required to change the
output voltage a fixed amount was dependent upon starting
wheel position. With the three wire system, the voltage
change per unit of rotation would be constant. The linear
voltage that represented wheel position could be analyzed
by comparators as is now being done, or could be converted
to a digital number that represented wheel position by an
analog to digital converter. Once the digital number was
obtained, the microprocessor could perform an analysis of
the recent corrections and alter the number that it

considered center to insure straight forward travel.

Microprocessor

If the aforementioned changes in the system sensors
were made, the VIP single board system would have to be
modified. The VIP has one eight bit input port and one
eight bit output port. This was adequate for the original
system but was not enough for the improved system. Only
one output port would be required, but at least four input
ports would be required. Two would supply the distance
count, one for the upper eight bits and the other for the

lower eight bits. A third input would be required to

86

supply the wheel position number. The last would supply
the remaining data such as learn mode select. The VIP
stored the total program in RAM. A practical system would
store the actual program in some form of ROM, most likely
EPROM. For these reasons, the next generation of the auto—
matic steering controller will need a different system.

The RCA 1802 has functioned well as a central processor
unit. Because it is a full CMOS microprocessor, it is well
adapted to the noisy tractor environment.

In the development process, several problems including
memory being changed were blamed on noise. In an attempt
to stop the supposed noise problem; a separate 12 volt
supply was used for the electronics. While this did lower
the noise content of the supply, this step might mot have
been necessary. The real problem was found to be that when
a switch mounted on the VIP board was moved, the board
would flex, shorting the five volt supply to the metal
enclosure. The regulator would go into its over current
protection state and return to normal when the short was
removed. In the meantime, some of the memory locations had
been changed. This problem was overcome by moving the
switch to the enclosure and by insulating underneath the
VIP board.

The improvements that were just listed will result in
more accurate and dependable control of the steering

function. Suggestions have been made for most of the major

87

system components. These suggestions, however, should not
detract from the fact that the system as described was
capable of mowing a lawn without an operator in direct
control. The microprocessor was capable of determining
its location on the grass cut edge and performing steering
corrections based upon that position. Ten specific
requirements of the system were stated in the system
general description. All ten of the requirements were met
by the original syStem. The concept used in the develop-
ment of the tractor was capable of automatic control of
the steering of a lawn tractor. Future work utilizing the
knowledge gained from this development, should result in

the development of a practical robotic tractor.

CHAPTER VIII

CONCLUSION

The robotic lawn tractor as described was capable of

mowing a rectangular lawn without operator assistance once

the programming out had been completed. The controller

performed all necessary functions to guide the tractor

along the desired pattern and was capable of halting

tractor and mower operations. Specific conclusions drawn

from the testing of the described controller are:

1)’

2)

3)

4)

5)

6)

The controller was able to maintain a minimum of
85 percent of the maximum mower cut width with a
lawn of uniform height and density;

In all testing of the ability of the controller to
guide the tractor along the grass cut edge, a
minimum of 99 percent of the grass was out under
automatic operation:

On dry grass, the controller could execute a 270
degree turn and reacquire the adjacent grass cut
edge in less than one meter of travel providing
the turn was started within 0.1 meters of the
corner;

The microprocesser was able to learn the dimensions
of the initial lawn with a maximum error of 0.3
meters;

The microprocessor reduced the length of the side of
the rectangular lawn by one meter after mowing the
side. The microprocessor was capable of continued
operation as the lawn plot was reduced in size;

The microprocessor was capable of halting forward
travel of the tractor and lawn mower operation:

88

89

7) A person unfamiliar with the machine would be
capable of operation with no more than five minutes
of instructions: '

8) The machine was capable of complete manual
operation.

LIST OF REFERENCES

LIST OF REFERENCES

Ambler, Harries, and Cox, Optical Ranging for Tractor
Guidance. 1980. ASAE Paper No. 80-1558.

Baxter, J., ed. Agricultural Engineer's Yearbook. 1981.
pp. 21 and 73.

Busse, W., Coenenberg, H., Feldman, F. and Crusinberry,
T. F., The First Serial Produced Automatic Steering
System for Corn Combines and Forage Harvesters.

1977. Grain and Forage Harvesting Conference
Procedings ASAE.

Gilmour, W. D., An Automatic Control System for Farm
Tractors. 1960. Journal of Agricultural
Engineering Research. 5:(4). pp. 418-432.

Grovum, M. A. and Zoerb, G. C., An Automatic Guidance
System for Farm Tractors. 1969. ASAE Paper No.

69-1170

Kanetoh, Yuuji, Driverless Combine Harvester. 1976.
Grain and Forage Harvesting Conference Proceedings
ASAE.

Kirk, T. G. and Krause, A. E., Swather Edge Guide Steering
Control System. 1975. ASAE Paper No. 75-1029.

Kirk, T. G., Zoerb, G. C. and Wilson, J. N., A Furrow-
Following Tractor Guidance System. 1976. ASAE
Paper No. 76-001.

Marley, C. F., Driverless Mower 'Sniffs' Underground
Guide Wire. Farm Show Magazine. 5-22-81.

Rushing, Karl, Developing the Driverless Tractor. 1971.
Agricultural Engineering. 52:260-262.

Schafer, R. L. and Young, R. E., An Automatic Guidance
SySEem for Tractors. 1978. ASAE Paper No.
76‘10680

Upchurch, B. L., Tennes, B. R. and Surbrook, T. G.,
Development of a Microprocessor Based Steering
Controller for an Over-the-row Apple Harvester.

90

91

1980. ASAE Paper No. 80-1556.

Young, R. E. and Schafer, R. L., Direct Digital Control

Young,

for Automatic Steering of Agricultural Tractors.
1974. IFAC Symposium on Automatic Control for
Agriculture, Saskatoon, Saskatchewan, Canada.

S. 0., Schafer, R. L. and Johnson, C. E., A
Microcomputer-Based Vehicle Guidance Controller.
1980. ASAE Paper No. 80-1557.

"IllTAIWANWT