Francisco Blanes, Gins Benet, Pascual Prez ,Jos E' Sim

1
MAP BUILDING IN AN AUTONOMOUS ROBOT USING
INFRARED SENSORS

• Francisco Blanes, Ginés Benet, Pascual Pérez
  ,José E. Simó
• Dep. Informática de Sistemas y Computadores.
  Universidad Politécnica de Valencia.
• Apdo. 22012, 46022 Valencia, Spain
• pblanes, gbenet, jsimo, pperez_at_disca.upv.es
• Dep. de Informática de Sistemas y Computadores.
• Universidad Politécnica de Valencia (Spain).

2
Project description
Mobile robot prototype
TAP97-1164-C03-03 (Mobile robot sensing)
Intelligent sensing Data fusion O.S. Kernel Real
time operation Reactive response
GV-C-CN-05-058-96 (Distributed systems)
TAP98-0333-C03-02 (Mobile robot sensing)

• Group background
• Real-time
• AI
• Intelligent control

Architecture definition
Objectives
To provide a general platform for research in
distributed architectures for real-time control
and mobile computing.
The platform is focused to operate in industrial
environments in a coordinated, controlled and
supervised autonomous way.
3
Prototype
4
Architecture
5
Main robot controller

• Embedded PC-board
• Network interface by radio (2 Mb/s)
• CAN bus manager unit
• Supervises and controls the whole robot.
• Operating systems RT-LINUX or Windows NT.
• Software
• Data sensor fusion and world modeling tasks

6
Motion controller

• 80C592-based board
• Controls two independent wheels that rotate
  around the same axis
• Optical encoders are used to obtain data about
  the estimated position of the robot (odometry).
• Two HCTL1100 circuits manage the power drivers
  for each motor.
• Speed control (PID)
• Position control (trapezoidal)
• Different modes of operation (Speed control,
  position control and autonomous trajectory
  tracking).

HCTL1100 Board
LMD18200T Driver
80C592 board
motor
Internal Bus
motor
HCTL1100 Board
LMD18200T Driver
CAN Bus
7
Sonar System

• 80C592 based board
• 2 x 8-bit A/D converter
• CAN bus subsystem
• Two sonar heads with different
• main lobe width.
• Controls the angular position of a rotating
• transducer by means of a stepper motor.
• Generates the ultrasonic waveforms to be supplied
  to the transducer.
• 4 Mb memory to process the echoes received from
  surrounding objects.
• Different modes of operation
• Distance measurement
• Local map building.
• Communicate the raw data to main processor.

10º
30º
Sonar module prototype
8
IR Sensors

• IR ring with 16 sensors
• Each sensor is composed with two LED and one
  photodiode
• The sensors are grouped in bundles of two sensors
• Only 16 milliseconds for a complete sequential
  scan

9
IR Sensors

• Sensor control based on 80C592 micro-controller
• Data sent trough CAN Bus in 8 bit format (4
  sensors in each message) or 10 bit (1 sensor in
  each message)

10
Model of the IR sensor
The sensor output can be modelled using the
following equation, as a function of the distance
x and the incidence angle ? with the target
surface

• ? and ? are the parameters of the model.
• The parameter? includes the radiant intensity
  of the IR emitters, the spectral sensitivity of
  the photodiode, the gain of the amplifier and the
  reflectance coefficient of the target ? i
• The parameter ? is the offset due to the ambient
  light
• ?o is constant, and ? can be obtained before
  every reading.
• Thus, the only parameter that characterizes an
  object is its reflectance coefficient ? i

Plot of the model for different values of ? i.
Normal values of ? i.
11
Model of the IR sensor
Response to different Canson coloured cards in
controlled tests

• The response depends mainly on texture of
  material rather than on colour
• Tests performed in the real environment showed
  typical values of ?i varying from 0.7 to 0.9

12
Incidence angle estimation

• Incidence angle is unknown a priori.
• If zero incidence angle is assumed, readings are
  oversestimated by a factor of cos ? .
• The value of ? can be obtained from the readings
  of a pair of transducers, as depicted in the
  figure.
• From the figure, the apparent angle ? can be
  obtained as
• tan ? (d1-d2)/L
• The exact value of ? can be derived solving this
  equation

• Instead, an approximated value can be obtained
  using

13
Distance treatment

• Distance estimation D could be used for grid map
  building but ...
• The estimation suffers uncertainty because colour
  and texture of the target
• We can model this uncertainty with a range of ?i
  values
• Common objets in the robot environment vary from
  an ?imax0.9 to ?imin0.7 (data obtained from
  tests)

14
Distance treatment

• The ?max produces a dmin estimation of the
  distance, ?min a dmax and hence a mean distance
  between them
• These distances define a sector of occupied cells
  

15
Sensor fusion

• Distance estimations are fused in pairs from the
  same couple
• Distance measurements are valid if the the
  incidence angle ? is between -45º,45º, over
  these values the error in the angle estimation
  implies errors bigger than 1.8cm
• If the distance value is valid, then dmax and
  dmin define a sector with a set of cells which
  occupancy value are calculated linearly

16
Map Building

• Local values of the cone are fused with the
  previous map ones
• A Bayesian approach has been used in this fusion
  process

p_occ_new
17
Tests and Results

• Tests were conduced in a real environment
  composed of brick walls, wood doors and metal
  doors.
• The goal of tests was to locate these objects in
  the scene
• Control system is based on threads and CAN
  communication to accomplish real-time
  restrictions
• The acquisition and fusion loop has a period of
  150 milliseconds
• Average speed of the robot is around 0.25m/s

18
Tests and Results

• Test 1 short corridor with metal cylinders

Scenario 1
cylinders
Resulting grid map 20x20 mm cell size
19
Tests and Results

• Test 2 Long corridor with doors

Odometric error 90º angle
Scenario 2
Resulting grid map 40x40 mm cell size
20
Tests and Results

• Test 3 end corridor with autonomous robot
  exploration

Resulting grid maps 20x20 mm cell size
75cm IR range
55cm IR range
21
Future Work

• New robot version
• Improvement of sensor configuration distance
  between sensor has been increased, thus a better
  incidence angle estimation can be obtained
• Ultrasonic and infrared sensor fusion could be
  used to improve distance estimation and to obtain
  the reflectance value

22
Conclusions

• A new IR sensor with large range has been
  presented
• The influence of the angle of incidence in the
  measurement is solved using pairs of estimations
• The uncertainty inherent to the colour and
  texture of objets is managed during the sensor
  fusion
• Good quality grid maps have been obtained using
  the sensor fusion approach