Development of a Human -driven tractor following a Robot System [627893]

Development of a Human -driven tractor following a Robot System

Chi Zhang . Liangliang Yang , Noboru Noguchi *

*Laboratory of Vehicle Robotics, Graduate School of Ag riculture,
Hokkaido University, Japan. (E -mail: [anonimizat])
Abstract: Automated agricultural equipment such as a robot tractor has been developed to solve the
problem of agriculture labor force shortage. A human -driven tractor following a robot system is useful
for saving time on a large -scale farm. In this kind of system, a r obot tractor does farm work while a
human -driven tractor is following and doing a different operation. The human operator can control the
robot through a controller. To monitor work conditions of the leading robot tractor, four cameras are
mounted on the r obot tractor, and images from the cameras are sent to the human -driven tractor through a
video transmission system. The human operator can check the surroundings of the robot from a monitor.
Also, a laser scanner is mounted in the front of the robot for sa fety. The results of an experiment using
the system showed that the precision of the robot tractor, w hich had an RMS error of about 0.03 m, was
better than that of an experienced tractor operator, for which the RMS error was about 0.04 m, while the
range o f lateral errors of the human -driven tractor improved from 0.25 m to 0.20 m by following the
robot .
Keywords : following system, robot tractor, human -driven tractor, laser scanner

1. INTRODUCTION
Increased and sustained agricultural productivity is neede d to
meet the globally increasing demands for food and energy.
The development of new agricultural machinery will enable
reduction in the cost of food production and provision of a
stable food supply. Over the past two decades, the age of the
overall agric ultural labour force has been increasing, while
the total number of farmers has been decreasing. According
to the Japanese Ministry of Agriculture, Forestry and
Fisheries, the agricultural workforce in 2010 was 2.6 million
people, a reduction of 2.2 millio n (45.6 percent) from the
workforce in 1990. Also, the average age of the agricultural
workforce increased from 59.1 years (1995) to 65.8 years
(2010). Although agricultural labour is decreasing and aging,
agricultural acreage per household in Japan increa sed from
1.83 ha (in 2007) to 2.07 ha (in 2012). In Hokkaido, the
agricultural acreage per household has increased by 15.5
percent in the past 6 years. In 2012, the average agricultural
acreage per household in Hokkaido reached 22.34 ha.
Automation of agri cultural machinery is considered to be one
of the most effective ways for improving productivity and
quality of various field operations (Noguchi et al., 2004).
Noguchi et al. (2001) developed a field robot based on
sensors including RTK -GPS and an inertia l measurement unit
(IMU). In their study, the robot could do farm work with an
error of only 0.05 m. Such a system can perform farm work
more precisely than an experienced human under straight line
conditions.
However, when a robot is used in an open field , a monitoring
system is required to ensure safety ( Noguchi et al., 2000). At
least one worker is needed to monitor the robot’s operation,
which is equal in efficiency to a single human drives a tractor (Noguchi et al., 2002), only to reduce the working st rength.
With improvements in robot technology and social needs,
researchers have become interested in a multi -robot system or
a human -driven tractor following a robot tractor system.
Noguchi et al. (2004) developed a master -slave robot system
to conduct fa rm work.
The objective of this study was to develop a robot tractor that
can perform farm work with a human -driven tractor. When a
robot tractor conducts an operation such as tillage, a human –
driven tractor just follow the robot to do proceeding work at
the same time. By following the robot, human driver can see
the status of robot tractor and improve driving accuracy. In
order to accomplish this objective, a remote controller for the
robot tractor was first developed. Then a communication
system that can receive a command from the remote
controller and send a command to the robot tractor was
developed. Fina lly, monitor system was used to assist human
drive.
The newly developed following system has several
advantages. Firstly, during the rainy season, farmers must
finish farm work in a limited time; otherwise, it will lead to
some disadvantages in crop yields . The following system can
reduce work time to improve such problems. Secondly,
compared with a single robot, the following system, in which
a single human operator can control two tractors at the same
time, reduces not only working time but also working
strength. Moreover, the following system helps in weed
control. Weeds generally begin to grow after tillage and have
an adverse impact on crops. In Hokkaido area, the average
farm was 22.34 ha, suppose the velocity of tractor was 1.5
m/s, thus each operatio n will take at least 18 h. However, the
following system can perform tillage and planting at the same
time, thus reducing the growth period between weeds and Proceedings of the 19th World Congress
The International Federation of Automatic Control
Cape Town, South Africa. August 24-29, 2014
978-3-902823-62-5/2014 © IFAC 11559

crop. Finally, compared with a large robot tractor, the
following system gain advantages over soil compaction, the
following system consisting of two middle -sized tractors
reduces damage to the crop and ground.

2. Experimental Equipment and Method
2.1 Equipment
2.1.1 Robot tractor
Fig. 1 shows the architecture model for the robot tractor. The
tractor used in this study was a Yanmar EG83, and the
specifications of the robot include steering control, a switch
for forward and backward movements, easy -change
transmission, a switch for power take off, hitch functions,
engine speed set, engine stop and b rake. A computer was
used to communicate with the tractor through a CAN bus. An
RTK -GPS with an embedded IMU (AGI -3, Topcon Co., Ltd,
Japan) was mounted on the top of the tractor, and the robot
tractor was equipped with a wireless video transmitter
(COSMOW AVE Co., Ltd, Japan), 4CH QUAD Processor
(Shenzhen Suntex Electronics Co., Ltd, China), four cameras
(DW INC Co., Ltd, Japan) , a laser scanner (HOKUYO
AUTOMATIC CO.,LTD, Japan) and a Bluetooth. Three
markers to assist the human operator were attached to th e
equipment on the robot tractor. One of the markers was in the
centre of the equipment, and the other two were aligned with
the wheels of the robot tractor. In this case, human can drive
the following tractor to follow the mark created by markers.

Fig.1 Equipment for robot tractor
Considering potential collisions during agricultural
operations, an obstacle avoidance system is an important
component in robotic applications. The remote video
transmission system included four cameras, video
transmitter/rece iver, 4ch quad processor and a monitor. Four
cameras were attached to the front and back of the robot
tractor on both left and right sides. The quad processor splits
the screen into 4 rectangles to show images from each of the
four cameras and then send th e images through a wireless
video transmitter. After receiving the images through the
wireless video receiver mounted on the manual tractor, the
images are displayed on a monitor. Thus, the human operator
can see all four sides of the robot tractor, enabli ng decisions
to be made in emergency situations. Fig. 2 shows a picture of
the video transmission system. If somebody comes near the robot tractor, the human operator can stop the robot to avoid
a collision. The human operator can also see the status of
equipment through the rear camera.

Fig.2 Wireless video transmission system
A laser scanner was mounted in the front of the robot tractor.
Once an object was detected in the predetermined zone, it
will send message to the robot to stop. The coordinates of
laser scanner is defined in polar coordinates system. The
coordinate of an object was transformed from laser
coordinates ( ρ,𝛉 polar coordinates) to vehicle coordinates ( x,
y Cartesian coordinates) , as in (1) .
x=ρ 𝗑 cosθ (1)
y=ρ 𝗑 sin𝛉

2.1.2 Human -driven tractor
A remote controller and a monitor are used to assist the
human operator. The following system includes two parts: a
remote controller and a remote video transmission syst em.
Fig. 3 (a) shows a picture of the remote controller and Fig. 3
(b) shows a flowchart of the algorithm of the remote
controller. The controller sends a command to the robot
tractor and receives feedback messages from the robot tractor.
The human operato r can change the parameters of a
command by using the relevant buttons. The monitor displays
the command and current robot tractor’s response status,
which helps the human operator to modify the tractor’s
command parameters.

(a) Picture of remote controller 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014
11560

(b) Flowchart of the algorithm of remote controller
Fig.3 Remote controller
In this study, the remote controller was developed based on
Windows API, and a windows platform was needed. The
human operator can control engine spee d, vehicle speed, PTO
on/off, hitch up/down and stop. The steering was controlled
by the robot itself without human involvement.
Communication software was developed to communicate
with the remote controller, and exchange messages with the
robot tractor ’s software. Two Bluetooth devices were used as
wireless communication tools. The devices were connected to
the remote controller through RS232. The effective working
distance of the communication was 150 m.
2.2 Method
2.2.1 Path plan
The width of the equip ment on the robot tractor is the same
as that on the human -driven tractor, and the overlap of the
path is zero. If the robot tractor goes to the neighbour path,
there will be a risk of collision of the equipment. To avoid
collision of the equipment on the tractors, the sequence of the
path order was disordered, with at least one path being
skipped from last path to the current path. For example, in a
six-path map, as shown in Fig. 4, the red numbers are the
work sequence of the robot tractor, and the order is
142536.
Fig.4 Path order of six paths
2.2.2 Headland turning method
A U-turning method was used in this study, as shown in Fig.
5. The turning steps are as follows.
Step 1: The robot goes straight forward from A to B an d then
turns at the maximum steering angle to point C.
Step 2: The robot calculates the distance between the current
path and the next path, which is w, and then decides the
distance between point C and point D, which is w-2r, where,
r is the minimum turni ng radius of the robot tractor. If w is
less than 2r, the robot will go backward to ensure a turning
radius.
Step 3: The tractor turns to the next path from point D, and
turning finishes at point F.

Fig.5 U-turning method
3. Experimental Results and Discussion
A field experiment was conducted on a farm in Hokkaido
University , Japan. The experiment was conducted to check

A B C D
E
F Field Edge
Next path Current path
r r w
1 2 3 4 6
6 5 ①
③
②
⑥
④
⑤
Start
Receive messages from
robot tractor
Read human control
buttons
Send command to robot
tractor
Display command and
robot tractor’s status 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014
11561

the response of the following system. In the experiment,
tillage was performed by the robot tractor and planti ng was
performed by the human -driven tractor, as shown in Fig. 6.
The robot carried a power harrow, and three markers were
mounted on the power harrow. The markers were aligned
with the left wheels, centre of the tractor and the right wheels,
respectively. The human -driven tractor was equipped with a
planter, which can do seeding and fertilizing. The widths of
the power harrow and planter were the same, and the working
width was 2.3 m. The overlap of the two paths was zero. And
the robot cannot get into the neighbour path because of
conflict between leader’s equipment and follower’s
equipment on headland and working path.

Fig.6 Working path of associating system
3.1 Performance of the tractors
For the experiment, we made a six-path map for the robot
tracto r. The sequence of the ten paths was reordered as
142536. Fig. 7 shows the map and trajectory of
the robot tractor. We used lateral error, which is the distance
between a predetermined path and current position of the
robot tractor, to evaluate the pe rformance of the robot tractor
and the human -driven tractor. Fig. 8 shows the lateral error of
the robot tractor.
5.2715 5.2716 5.2717 5.2718 5.2719 5.272
x 105476934476936476938476940476942476944
UTM-Easting [m]UTM-Northing [m]

Trajectory
Map

Fig.7 Map and t rajectory of robot tractor
0 100 200 300 400 500-15-10-5051015
Time [s]Lateral error [cm]Path1 Path4 Path2 Path6 Path3 Path5
(a) Lateral error of 6 paths
100 120 140 160 180 200-8-6-4-202468
Time [s]Lateral error [cm]

(b) Zoom in path4
Fig.8 Lateral error of robot tractor
The maximum error was 0.16 m, and the minimum error was
-0.17 m. The average RMS of the lateral error of all 6 paths
was 0.03 m. The average of absolute value of lateral error
was 0.02 m, and t he average lateral error was -0.003 m,
which is an acceptable value.
Fig. 9 shows turning accuracy of the robot tractor. Turning
accuracy is lateral error of the beginning of the working path,
from point E to point F, as shown in Fig. 5. The maximum,
minimum, average and RMS of t he turning lateral error were
0.12 m, -0.14 m, 0.01 m and 0.08 m, respectively. The
average of absolute value of lateral error was 0.08 m. At the
end of the turning part, the average lateral error of all 6 paths
was -0.01 m. The turning accuracy was 0.06 m lower than
that of the working part, but it is also an a cceptable value.
-15-10-5051015Lateral error [cm]
3 -> 6 5 -> 3 2 -> 5 4 -> 2 1 -> 4

Fig.9 Turning accuracy 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014
11562

Fig. 1 0 shows the effectiveness of the newly developed
following system in terms of accuracy of the human -driven
tractor. As described above, markers were mounted on the
robot tractor, and the human operator follow ed the track
generated by the markers. The average velocity of human –
drive alone and human drive following robot was different.
The velocity of human -drive alone was 0.9 m/s, that 0.1 m/s
faster than that of following the robot. Fig. 1 0 (a) shows the
lateral error of human -driven tractor without following the
robot. The maximum and average of the lateral error of all 6
paths were 0.10 m and 0.02 m, respectively. And the average
of absolute value of lateral error was 0.05 m. Fig. 1 0 (b)
shows the performance of the human -driven tractor following
the robot tractor. The maximum and average of the lateral
error of all 6 paths were 0.08 m and -0.007 m, respectively.
And the average of absolute value of lateral error was 0.03 m.
By following the robot tractor, the precision of the human –
driven tractor improved by forty percent, from 0.05 m to 0.03
m.
0 100 200 300 400-15-10-5051015
Time [s]Lateral error [cm]Path6 Path3 Path5 Path2 Path4 Path1

(a) Human -drive alone
0 100 200 300 400-15-10-5051015
Time [s]Lateral error [cm]Path1 Path4 Path2 Path5 Path3 Path6

(b) Human -drive following robot
Fig.10 Lateral error comparison of human -driven tractor
The human driver had over 30 years of driving experience .
The results showed that the range of lateral errors of the
human -driven tractor improved by twenty percent, from 0.25
m to 0.20 m, by following a leading robot. Also, it will
reduce human’s working strength by following the robot
tractor.
3.2 Performance of the remote controller In this study, the newly developed following system allowed
the operator on the following tractor to control the engine
speed and velocity of the robot tractor while the robot was
tracking a straight line. Fig. 1 1 shows the engine speed in
response to the command from the human operator. The
engine speed was used to check the current situation of the
robot tractor. If the speed of the robot and PTO rotation rate
were so high that the engine cannot satisfy, the human driver
can eith er decrease the velocity or increase the engine speed.
0 100 200 300 400 5008001000120014001600180020002200
Time [s]Engine Speed [rpm]

Response
Command

Fig.11 Engine speed response of robot tractor
Fig. 12 shows vehicle speed in response to the command
from the human driver. As shown in Fig. 1 2 (b), the time
delay was about 1.8 s when the velocity was changed . The
time delay includes delay of remote controller, delay of
communication software, delay of robot tractor ’s software
and delay of tractor ’s ECU. From the logging data, the delay
of remote controller and communication software can be
calculate d. The delay of remote controller was 0.2s, and
delay of communication software was 0.3s. The robot
tractor ’s delay was 1.3s. In this experiment, the velocity of
the robot tractor was 0.8 m/s, which means it continues work
about 0.72 m after the human oper ator has pressed the stop
button. It is concluded that this distance is acceptable for
stopping the robot tractor in an emergency situation. In
addition, restarting the robot took 2.6 s. A function was
added to the robot system by which the robot checks th at one
second has passed to ensure safety before it restarts after
stopping.
0 100 200 300 400 50000.511.52
Time [s]Vehicle Speed [m/s]

Response
Command

(a) Vehicle speed response of robot tractor 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014
11563

110 120 130 140 150 160 1700.40.60.81
Time [s]Vehicle Speed [m/s]

Response
Command

(b) Zoom in area of (a)
Fig.12 Vehicle speed response of robot tractor
4. Conclusions
In this study, a robot tractor following with a human -driven
tractor was developed to improve agricultural work efficiency
and safety for robot operations.
A robot tractor and a human -driven tractor were used in this
study. The average lateral error was less than 0.03 m. The
human operator just followed a marked line created by the
leading robot tractor. The human operator used a controller to
control the robot, and equipment for observing the robot’s
surroundings by a wireless video transmission system was
installed. One human operator can maneuver two tractors,
and the efficiency of farm work is thus improved. Also, by
following the robot tractor, the human operator can drive on a
predetermined path more accurately, from 0.06 m to 0.03 m.
The range of the human operator’s lateral errors improved
from 0. 25 m to 0. 20 m. The accuracy of human -driven tractor
was high enough for agriculture work.
REFERENCES
Noguchi, N., Ishii, K., T., H., 1997. Development of an
Agricultural Mobile Robot using a Geomagnetic
Direction Sensor and Image Sensors. Journal of
Agricu ltural Engineering Research 67(1997), 1 -15.
Noguchi, N., Reid, J.F., 2000. Engineering challenges in
agricultural mobile robot towards information agriculture.
In: Proceedings of the XIV Memorial CIGR World
Congress, pp. 147 -154.
Noguchi, N., Jeff, W., J.R ., Q.Z., 2004. Development of a
master -slave robot system for farm operations.
Computers and Electronics in Agriculture 44(2004), 1 -19.
Barawid Jr, O.C; Mizushima, A .; Ishii, K.; Noguchi, N., 2007.
Development of an Autonomous Navigation System
using a Two -dimensional Laser Scanner in an Orchard
Application. Biosystems Engineering 96 (2007),139 -149.
Olivier S., Olivier G., 2009. A cooperative multi -robot
architecture for moving a paralyzed robot. Mechatronics
19(2009) 463 -470.
Rui R., Jorge D., A.C., 2005. Cooperative multi -robot
systems: A study of vision -based 3 -D mapping using
information theory. Robotics and Autonomous Systems
53 (2005) 282 -311. 19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014
11564