Transcript Slide 1

Autonomous Navigation in Forest Environment
GOALS FOR THE PILOT PROJECT:
THE VISION OF FORESTRY 2015?
A prototype for a forest vehicle, capable of:
 Learning a path run by a human operator
 Autonomously tracking the path while accounting for
 Unplanned deviations from the path
 Obstacles such as trees, human beings and animals
Manned Harvester
 Hard to make autonomous
 The driver supervises one or several
autonomous wood shuttles
Why unmanned?
 Lower labor costs
 Lower weight 
- Less ground damages
- Less emissions
Unmanned
Wood shuttle
Operator program:
• Path learning
• Path tracking
• Tele operation
Takes the trees out of the forest
Sensor data:
GPS: 63.53.341N 20.19.247E
Compass: 213.5 deg.
Radar: 32 deg.
2.17 meter
276 deg.
0.45 meter
Control commands:
Turn(3.1)
Speed(0.36)
Halt
Switchboard
Modified
Pioneer AT2
Simulator
A NEW PATH-TRACKING ALGORITHM:
Follow-the-Past
v’
Safe speed
vs
Φα
Sonar
Laser
Radar
Occupancy grid
Φsum
Avoid
obstacles
(VFH+)
vt
s
Modified Valmet 830
The work has been
conducted on three
different target
machines, each with
increased complexity.
This approach greatly
simplifies the research
and development of
both hardware and
software.
TESTS WITH VALMET 830:
The resulting set steering angle is the sum of Φα, Φβ and Φγ.
When the distance to the path has been reduced close to
zero, the set steering angle essentially equals Φγ, i.e. the forest
machine tries to mimic the recorded steering commands of the
human driver.
The forest machine
initially started 15
meters away from the
recorded path:
vo
turn
(x’, y’)
Recorded path
(x, y)
GPS/INS

’
Φ’
Move towards
Φα
path
w1
Turn towards
the recorded
orientation ’
Φβ
Mimic the
recorded
steering Φ’
Φγ
w2
+
w4
s
Φt
Well known problem of cutting the
corners with traditional algorithms
w3
Follow the Past consists of three reactive behaviors,
which are then fused into a single steering command:
Φα : Move towards path
Φβ : Turn towards the recorded orientation
Φγ : Mimic the recorded steering
(a) Follow the Past
(b) Follow the Carrot
SYSTEM ARCHITECTURE:
SENSOR FUSION:
Mobile computer
Win XP Java
- Hardware interface
to sensors and actuators
- Low-level control loop
- Occupancy grid
- Laser odometry
- Communication
Sensors
For localization:
GPS/GLONASS
Gyro/Compass
For obstacle detection:
24 GHz radar
Laser scanner
Engine rpm
Steering angle
Since the GPS position may be incorrect when
the vehicle tilts, the fused position is corrected
with respect to vehicle roll and pitch.
Thomas Hellström
[email protected]
Ola Ringdahl
[email protected]
Position from Laser odometry is fused with other
sensors to get a better estimate of the position
when the GPS does not work. The position error
from the laser is less than 0.5 m in this example.
Department of
Computing Science
Umeå University
Sweden
(c) Pure Pursuit
WLAN
communication
Remote computer
Win XP Matlab
- User interface
- Path learning
- High-level control loop
for Path tracking
- Data analysis routines
under development
- Communication
DGPS base station
Actuators
Steering
Throttle pedal
Brake
Overview of the architecture of the developed
system. The high- and low-level parts are split
between two computers. The forest machine shown
is a vision of what a future autonomous shuttle could
look like.
Project sponsored by Carl Tryggers Stiftelse,
Kempe Foundations, Komatsu, Land Systems
Hägglunds, LKAB, Umeå University, VINNOVA