Abstract
Laser obstacle sensors are undergoing development at a very fast pace.These are the most accurate among distance sensors but their high costs on account of not so easily available materials used in their construc- tion delayed their deployment for commercial applications. This led to the creation of a low cost laser sensor called Revo LDS which is made using ma- terials easily available o the shelf and cost under $ 30. Also discussed is the practical implementation of such a sensor in vehicle navigation.

[attachment=7220]

Contents
1 INTRODUCTION 1
2 LASER DISTANCE SENSOR 2
2.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Solution - The Revo LDS . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 The Revo Design . . . . . . . . . . . . . . . . . . . . . 4
3 SINGLE POINT DISTANCE MODULE 5
3.1 Triangulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Single Point Module Design Criteria . . . . . . . . . . . . . . . 6
3.3 Module Calibration . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4 Laser Dot Localization . . . . . . . . . . . . . . . . . . . . . . 9
3.5 Ambient Light Rejection . . . . . . . . . . . . . . . . . . . . . 10
3.6 Electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.7 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4 SCANNING AND SYNCHRONISATION 13
4.1 Rotational Scanning . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Angular Synchronization . . . . . . . . . . . . . . . . . . . . . 14
4.3 Durability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
i
5 APPLICATION 15
5.1 Unmanned Ground Vehicle System . . . . . . . . . . . . . . . 15
5.1.1 Obstacle Classication . . . . . . . . . . . . . . . . . . 16
5.1.2 Obstacle detection system . . . . . . . . . . . . . . . . 17
5.1.3 Data clustering . . . . . . . . . . . . . . . . . . . . . . 18
5.1.4 Local map generation . . . . . . . . . . . . . . . . . . . 19
5.1.5 Terrain estimation . . . . . . . . . . . . . . . . . . . . 19
6 CONCLUSION 21
ii
Chapter 1
INTRODUCTION
Obstacle detection eld is a broad one and development of obstacle sen-
sors is going on at a fast pace . These sensors nd a lot of applications in
elds ranging from military operations to GPS data acquisition.
The most important among these are LASER sensors . The advan-
tage with Laser sensors is that laser is a highly directional beam and can
travel good distances(depending on the application at hand and type of laser
employed) without any signicant spreading. This helps in designing sys-
tems with high SNRs. As a result these sensors can be employed in entirely
dierent scenarios like tracking fast moving objects and also servo as input
devices for future generation of computers.
In most scenarios instead of deploying a single sensor to perform the
desired operation , system designers often tend to use combination of sensors
to complement each other akin to human senses working in tandem. The
data from dierent sensors are obtained and sensor fusion algorithms are
employed to nd a better estimate of the measured parameters. Due to the
high accuracy in measurements given by these sensors they nd applications
in robotics.
1
Chapter 2
LASER DISTANCE SENSOR
2.1 The Problem
One of the most common tasks for mobile robots is to make a map
and navigate in an environment. To do so, the robot needs to sense its
environment in an ecient way, looking out to some distance to nd obstacles
and build a map that is useful for performing tasks such as vacuuming or
delivery.
While there are many sensors that could be used, laser distance sen-
sors are currently the standard sensors in indoor and outdoor mobile robots.
The main reason is the utility of the data: an LDS returns distance to objects
in its eld of view, unlike (for example) vision sensors, which need compli-
cated and error-prone processing before distances are measured. And unlike
other distance sensors such as sonars or IR sensors, an LDS is capable of ne
angular and distance resolution, realtime behavior (hundreds or thousands
of point measurements per second), and low false positive and negative rates.
While LDS devices are ubiquitous in research robotics,their high cost
has kept them from appearing in consumer robotics such as robot
oor clean-
2
ers. The Electrolux Trilobite, one of the only cleaners to make a map, relies
on sonar sensors . The barrier to using LDS technology is the cost. The two
most common devices, the SICK LMS 200 and the Hokuyo URG-04LX ,cost
an order of magnitude more than the simplest robot cleaners.
2.2 Solution - The Revo LDS
The unavailability of low cost laser sensors led to good research and the
outcome of which was The Revo Laser distance sensor. Figure 2.1 shows the
protype Revo with its cover removed.
Figure 2.1: Revo LDS. Approximate width is 10cm. Round carrier spins,holds optical module with
laser dot module,imager, and lens.
It has the following characteristics.
 Eye-safe(Class I or II).
 Works under standard indoor lighting conditions, and some outdoor
conditions.
 Measures a full 3600 planar scan.
3
 Has a range from 0.2m to 6m.
 High resolution:range error < 3 cm at 6m,angular resolution 10.
 4000 readings per second.
 small size,low power(<2W).
 Standard , commercially-available components.
 Low cost :$30 cost.
These characteristics make the Revo suitable for consumer products,
and open the way for high-performance, low-cost mobile robots.
2.2.1 The Revo Design
A compact, rigid point-beam triangulation module incorporating laser,
imager, and electronics. With a low-cost CMOS imager and a DSP for sub-
pixel interpolation, we get good range resolution out to 6m with a 5 cm
baseline, at a 4 KHz rate. The key insight to the Revo is that high precision
is possible with a small baseline, because of the digital image sensor.
Module rotation to achieve a 360 FOV. Rather than using mirrors
to manipulate the beam, the Revo revolves the optical assembly to point the
beam around a full circle.
No other current device satised their requirements. They brie
y
reviewed several relevant competing LDS technologies that use triangulation.
4
Chapter 3
SINGLE POINT DISTANCE
MODULE
The Revo relies on an innovative laser point sensor module that works
on the triangulation principle, using a laser point beam and a digital image
sensor, separated by a small baseline. The module incorporates laser, sensor,
optics, and computation in a small, rigid package . It is slightly larger than
current IR distance sensors , but has much better accuracy and speed.
3.1 Triangulation
All single-point scanning sensors, such as the SICK and Hokuyo devices,
use mirrors to scan the point sensor. These devices are time-of-
ight distance
sensors: they measure the time it takes for light to travel to an object and be
re
ected. An alternative technology is triangulation: distance to an object is
measured by the angle of the re
ected light. Fig2 shows the basic geometry
of triangulation. A laser produces a small point of light, which re
ects o
an object and onto the image plane of the camera. An ideal pinhole camera
5
is oriented so that the laser beam is parallel to the ray through the center
of focus to the edge of the image. This gives a distance measurement from
innity (at one edge of the image) to the distance qmin (at the other edge).
From similar triangles, the perpendicular distance to the object is
q =
fs
x
(3.1)
The distance along the laser ray depends also on the angle of the laser
with respect to the image axis:
d =
q
sin
(3.2)
These equations show the hyperbolic relationship between image dis-
tance and object distance that is a property of triangulation. This nonlinear
relationship poses problems for determining longer distances the range sen-
sitivity dq/dx grows quadratically with distance.
(3.3)
For example, if a 1-pixel image displacement corresponds to a 1 cm
distance displacement at 1m, then it corresponds to a 4 cm displacement at
2 m.
3.2 Single Point Module Design Criteria
The criteria for minimum distance (from Eq 3.1) and range resolution
(Eq. 3.3) pull in opposite directions: a small fs product gives a small qmin,
a large fs has good range resolution.
The relative weight of fs is determined by the image sensor, so they
rst decided on it. The image sensor should have a short exposure time to
improve ambient light rejection , and a large number of pixels for resolution
6
Figure 3.1: Triangulation geometry from similar triangles.
of x.They chose a global-shutter CMOS sensor with 752 pixels of resolution
and a minimum shutter time of 35s. Each pixel is 6m, and they expected
it to be able to resolve the laser dot to within 0.1 pixel or better.
With these parameters, they plotted the eect of fs on range reso-
lution and min distance (Figure 3.2). If the min distance is to be 20cm or
less, fs should be 900 or less. If the range resolution is to be 30mm or less
at 6m, the fs product should be greater than 700. They picked 800 as the
sweet spot for the device.
The product fs = 800 can be achieved in dierent ways,but the bias
is towards a compact baseline, while keeping the focal length reasonable
(larger focal lengths demand longer lenses). With a baseline of 50mm, the
focal length is 16mm, and they chose this combination.
7
Figure 3.2: Min distance and range resolution relative to fs. The vertical line is the sweet spot.
3.3 Module Calibration
The total error of the device is a function of the device parameters, the
error in dot resolution, and the calibration of the device. Calibration here
refers to all the misalignments that could aect the ideal operation of the
device. Because they used low-cost optical components, the design must
account for major inaccuracies. The main ones are laser pointing angle, lens
pointing angle, and lens distortion.
 Laser pointing angle Laser must point vertically in a plane parallel to
the base of the device and point horizantally at 80 angle towards the
principle ray of the camera.
 Lens pointing angle The Laser beam and lens principal ray need not be
in the same plane.In calibration they searched for the best horizontal
scanline that corresponds to the laser beam at all distances.
 Lens distortion. The distortion at edge of the eld will be atleast a few
8
Figure 3.3: 1/x calibration curve
percent,even when optimising for a single wavelength.
They used a two stage calibration process.
 Localize the laser dot image to subpixel accuracy.
 For a set of readings of known distances,t the ideal curve of Eq.1
,weighting distant readings more heavily.
3.4 Laser Dot Localization
To reduce errors at larger distances, the image of the laser dot must be
localized to subpixel precision. They used a simple centroid algorithm for
localization. First, the rows in 10-pixel horizontal band are summed. The
resultant line image is then dierentiated and smoothed, and the center of
the dot is found using the maximum value. Finally, the centroid is calculated
9
as

I(i):i
I(i)
(3.4)
At short distances, the image of the dot is tens of pixels wide, and
large STDs are tolerated. At longer distances,the dot image becomes only
a few pixels wide, and sampling eects can be important. The anomalous
readings at 2 m and 4 m are probably a result of the dot being near a pixel
boundary, a well-known eect in nding the center of a dot . They plan to
investigate matched lter methods for better subpixel localization . Even
with the centroid method, localization is at 0.2 pixels or better for longer
distances.
3.5 Ambient Light Rejection
In most environments, the image of the laser dot is corrupted by am-
bient light. Two techniques for rejecting this interference are temporal and
wavelength ltering. They chose a visible red wavelength (650 nm) for the
laser, because it yields slightly higher laser output for eyesafe use, has better
imager response, and is easier to debug and calibrate than IR wavelengths.
A 20 nm bandpass lter reduces the ambient light
ux.
3.6 Electronics
The block diagram of Figure 3.4 shows the main electronic components.
The CMOS imager has integrated timing and control, and requires only a
frame pulse to start exposure and subsequent readout of 10 rows; the same
pulse starts the laser output. The processor, a DSP, streams the image data
directly into internal memory, where it is processed to nd the dot centroid
10
and map the centroid position to distance. The only external memory cali-
bration data.
Figure 3.4: Block Diagram of main components.
All the main components t on a small PC board attached to the lens
module . The module uses less than 1W of power in normal operation.Exposure
and readout occur sequentially, while processing is performed in parallel with
these operations. The primary limitation on speed is the time taken to read
out 10 lines. With on-imager binning of lines, it is possible to perform an
expose-process-readout cycle in under 0.25 ms, for a read rate of 4000 dis-
tance points per second.
3.7 Performance
They tested the LDS single-point module in two ways:
1. Error vs. distance for a newly-calibrated module, using white targets
(>90% re
ectance).
2. Error vs. distance for 10% re
ectance.
Figure 3.5 shows laser dot localization errors for targets with 10% and 90%
11
Figure 3.5: Rigid frame for the sensor module.Overall length is 7cm
re
ectance. These errors are random errors that arise from trying to localize
the position of the laser dot on the imager to sub-pixel precision.
Figure 3.6: Distance errors
12
Chapter 4
SCANNING AND
SYNCHRONISATION
To increase the eld-of-view of a single-point distance sensor, it must be
scanned. The typical scanning conguration for triangulation sensors users
mirrors to de
ect the laser beam and return re
ections to the image sensor.
Such an arrangement is inherently bulky and dicult to calibrate, requiring
precise positioning between mirrors, imager, and laser. It is also dicult to
achieve full scanning coverage typically coverage is 1800 or less.
By contrast, the Revo module is small and rigid enough to be me-
chanically scanned. In the current device, the module is rotated in a plane,
generating a full planar scan at up to 10 Hz. This unique mechanical arrange-
ment, without costly mirrors and consequent alignment problems, enables
the Revo to function reliably, while keeping manufacturing costs low. Other
arrangements of the module are also possible, e.g., a full 3D scan could be
generated by having the module measure not just a single point, but a set of
points, or a laser.
13
4.1 Rotational Scanning
The Revo module is mounted on a bearing and spun around an axis
midway between the laser and the imager . As the module is rotated, the
laser is pulsed and a reading is taken at 1 degree resolution. At 10 Hz
rotation rate, this is 3600 readings/second, below the maximum rate of 4000
readings/sec.
Power for the laser module is supplied through a 2-wire slip ring
on the rotation center. Communication to the module is via a short-range
radio frequency modem, at 115 Kbaud, sucient to send 2-byte data for each
reading.
4.2 Angular Synchronization
The Revo incorporates a low-resolution optical encoder on the rotating
module. A xed radial black-and-white pattern is read by two re
ective
sensors on the module. One of the sensors reads an index mark to give the
nominal heading of the Revo, while the other reads a 30 cpr pattern for timing
the exposure of the laser and imager. Using this technique, the angular
displacement of the laser readings is relatively insensitive to variations in
motor speed, allowing for cheaper motors and relaxed motor control.
4.3 Durability
Mechanical scanning of the optics module raises issues of durability. The
center slip is rated at least 1000 hours, good for 3 years of use at 1 hour/day.
Lifetime of the drive should be determined by the motor lifetime.
14
Chapter 5
APPLICATION
In 2004, the Defense Advanced Research Projects Agency (DARPA)
started the DARPA Grand Challenge to encourage development of Unmanned
Ground Vehicles (UGVs). After there were no winners in the 2004 Grand
Challenge they held the 2005 DARPA Grand Challenge. In 2007 they held
the DARPA Urban Challenge. All three events sparked much interest and
there were many competitors. In all three challenges, the Laser Measurement
Scanner (LMS) was an essential sensor, used by many competitors. In this
section, the estimation of various driving environments using multiple laser
scanners is discussed.
5.1 Unmanned Ground Vehicle System
The UGV was built using a popular Sports Utility Vehicle from Hyundai
Motor Company called the SantaFe. The system is divided into 5 main bode
systems: the vehicle control system, the navigation system, the obstacle
detecting system, the path planning system and the arbiter system. Each
node system consists and several hardware and software elements. The node
15
Figure 5.1: UGV system conguration
systems are connected through a UDP network. Software is developed using
LabVIEW which is developed by National Instruments.
5.1.1 Obstacle Classication
Obstacle detection is the most important part of terrain estimation.
Laser scanners provide range values to possible obstacles and a risk judgment
is assigned by an algorithm. Each obstacle is judged to be either a threat or
a non threat to the UGV. Threat obstacles are further divided into static or
dynamic obstacles. Obstacles are categorized into three general classes
1. Obstacles not on the road or drivable surface (trees, streetlights, people)
2. Obstacles on the road or drivable surface
3. The curb
Class I obstacles, which include trees, street lamps and pedestrians,
are located outside of the curb. Class I obstacles are generally classied as
non threats and do not in
uence the UGV behavior. If an obstacle suddenly
16
enters the roadway from o of the roadway it will be immediately classied
as a Class II obstacle it crosses the curb. As such, the size and form of Class I
obstacles are not taken in consideration since they are outside of the interest
area and no processing is required. Class II obstacles include anything on
the road or drivable surface area, such as pedestrians crossing the road or
other vehicles. Speed bumps are also considered Class II obstacles since they
have a direct eect on UGV navigation. Generally, speed bumps are detected
using cameras because it is dicult to dierentiate speed bumps from the
ground. If a speed bump exists in the road then an error is introduced into
the least-squares method used. As such, detection of speed bumps using the
laser scanner is attempted to minimize the introduced error. Most obstacles
on the roadway are the obstacles that the UGV must avoid and can be either
dynamic or static obstacles. In the applied system, all obstacles are assumed
to be static since the laser scanners measure continuously even if the obstacles
move. If the detection system always recognizes a moving obstacle as a static
obstacle, it is not necessary to predict the obstacles path. Except in sudden
situations, treating an obstacle as static is enough to dene the obstacle.
5.1.2 Obstacle detection system
The obstacle detection system is one of the node systems within the
whole system. Its main purpose is terrain estimation. The data from each
laser scanner is used by one and only one software component. Each compo-
nent processes the data and all the results are fused by a separate software
component which also performs the terrain estimation.

hi ...........Ashok,
can you send the ppt and full report on this topic??????
in advnc thanks.....

for full report, please download the attached file.