asas t PowerPoint

1
Technical University of Crete Dept. of Production
Engineering Management Intelligent Systems
Robotics Laboratory
Mobile Robots Autonomous Navigation Control
Architectures
2
Research Team

• Faculty
• Kimon Valavanis (Professor, and Director)
• Nikos Tsourveloudis (Assistant Professor)
• Graduate Students
• Lefteris Doitsidis (M.Sc. candidate)
• Undergraduate Students A group of 8 conduct
  their diploma thesis in the lab.

3

• Presentation Outline
• Steering Configurations
• Control Architectures
• Vehicle Navigation

4
Steering Configurations
Different kinds of steering configuration are
proposed

• Single Axis Steering
• Double Axis Steering
• Skid Steering

5
Single Axis Steering
d1outside wheel heading d2inside wheel
heading Rvehicle radius Lvehicle length
Bvehicle width Oturn center location
6
Double Axis Steering
d1outside wheel heading d2inside wheel
heading Rvehicle radius Lvehicle length
Bvehicle width Oturn center location
7
Skid Steering
vooutside wheel velocity v1inside wheel
velocity Vvehicle velocity Ozvehicle angular
velocity Lvehicle length Bvehicle width
Rvehicle turn radius
8
Key issues in Robotics Teams

Several approaches has been proposed in order to
categorize robotics teams
Premvuti and Yuta 1990 Yuta 1993 proposed

• Active or non-active cooperation
• Level of independence
• Types of communication
• explicit
• implicit

9
Dubek et al. proposed a more complete taxonomy
based on

• Team size
• Communication range
• Communication topology
• Communication bandwidth
• Team reconfigurability
• Team unit processing ability
• Team composition

10
Cao et al. proposed another architecture based in
four principal research axis

• Architecture
• Differentiation of agents
• Communication structures
• Via environment
• Via sensing
• Via communication
• Models of other agents intentions,capabilities,st
  ates, or beliefs

11

• There exist several proposed solutions to the
  problem of autonomous mobile robot navigation in
  2-D uncertain environments. They include
• Roadmaps (visibility graphs, Voronoi diagrams,
  freeway net)
• Exact and approximate cell decomposition
• Artificial Potential fields
• Fuzzy Logic
• Evolutionary Algorithms
• Methods combining Fuzzy Logic with Genetic
  Algorithms
• Fuzzy Logic with Electrostatic Potential Fields
• .to mention just a few.

12

• Autonomous Vehicle Navigation Utilizing
  Electrostatic Potential Fields and Fuzzy Logic,
  IEEE Transactions on Robotics and Automation,
  Vol. 17, No. 4, pp 490-497, 2001.
• Mobile Robot Navigation in 2-D Dynamic
  Environments Using Electrostatic Potential
  Fields, IEEE Transactions on Systems, Man and
  Cybernetics, Part A, Vol. 30, pp. 187-197, 2000.
• Fuzzy Logic Based Skid Steering Vehicle
  Navigation, IEEE International Conference on
  Robotics and Automation, Washington D.C, 2002.

13
Autonomous mobile robot navigation based on fuzzy
control(our approach)

• Both navigation and obstacle avoidance
• Control Architecture 2-level Fuzzy Logic Based
  Controller
• In the first layer the collision possibilities
  in the four main directions (front, back, left,
  right) are extracted (COMMON TO BOTH TYPES OF
  MOBILE ROBOTS)
• In the second layer the translational and
  rotational speed are decided based, on the
  inputs (collision possibilities and angle error)
• Application NOMAD200, NOMAD150, RWI ATRV-Mini

14
A novel approach to solving the Autonomous Mobile
Robot (AMR) navigation problem in 2-D dynamic
environments, is by combining the Electrostatic
Potential Field (EPF) path planner with a
two-layered Fuzzy Logic inference engine.

• Tasks are performed by
• Object Detection
• Localization
• Path planning and collision avoidance

15
EPF and Fuzzy Logic inference engine operate in
tandem
16
Given a 2-d environment the EPF plans the initial
trajectory
Once the object detection module detects through
sensors readings a high collision possibility,
it forces the motion control module to forget
the initial EPF path, and take corrective actions
in terms of robot steering and robot speed to
avoid the collision until the collision
possibility becomes low or not possible
Then the motion control module takes in account
the initial trajectory as computed at this time
instant by EPF planner
The EPF planner is re-invoked every time the
environment map is updated
17
The Natural Potential Field
The algorithm to create a natural potential field
follows three steps
Create an occupancy map of the environment
Create the resistor network
Solve the resistor network to obtain potential
field
18
Similarity to Dynamic Programming
The Dynamic Programming (DP) based approach to
the shortest path finding problem is divided into
the sub-problems of finding the next step plus
finding the rest of the path with the total cost
given by the general expression


The EPF based solution may also solve the problem
in a similar manner. Ohms Law determines the
electric current of a path as the product of the
potential and the conductance
The overall conductance of the path can be split
accordingly
Both DP and EPF algorithms require knowledge of
the immediate next step, as well as complete
knowledge of the rest of the path
Dynamic programming algorithms, backtracking
algorithms, and the EPF algorithm all guarantee a
shortest path approach
Both the DP and backtracking algorithms operate
with a complexity magnitude of O(n2), n is the
number of nodes in the network
19
The navigation control for the AMR is generated
through a two layer fuzzy inference engine
The first layer is responsible for obstacle
detection and performs sensor data fusion from
sonar readings
The second layer receives the output of the first
inference, representing the immediate collision
possibilities, as well the output of the
artificial potential field and the speed of the
robot as input and generates the control
20
(No Transcript)
21
Obstacle detection control motion modules
22
(No Transcript)
23
Experimental Results
Simulated dynamic environment test case the EPF
approach
24
Experimental Results
Simulated dynamic environment test case the FL
approach
25
Experimental Results
Simulated dynamic environment test case the
combined EPF/FL approach
26
Experimental Results
Experimental test case for the EPF approach
27
Experimental Results
Experimental test case for the FL approach
28
Experimental Results
Experimental test case for the EPF/FL approach
29
(No Transcript)
30
Skid steering vehicle motion differs from
explicit steering vehicle motion in the way the
skid steering vehicle turns. The wheels rotation
is limited around one axis and the lack of
steering wheel results in navigation determined
by the speed change in either side of the skid
steering vehicle.
Geometric Configuration of the vehicle in the X-Y
plane
at Heading Angle w Vehicle Width ?
Sense of Rotation S1 , S2 Speeds in either
side of the vehicle
31
Sensors Grouping
The sensors are grouped in pairs. Twelve sonar
sensor groups Ai, i1,,12, have been enumerated
as shown in the figure. The minimum of the
readings of each pair was considered as a
distance measure from the (potential) obstacle.
32
Again.. a two-layer fuzzy logic controller has
been designed for 2-D autonomous navigation of a
skid steering vehicle in an obstacle filled
environment.
First Layer The controller provides a model for
multiple sonar sensor input fusion and it is
composed of four individual controllers, each
calculating a collision possibility in front,
back, left and right directions of movement.
Second Layer Consists of the main controller
that performs real-time collision avoidance while
calculating the updated course to be followed by
the vehicle.
33
First Layer
The form of each first layer individual fuzzy
controller, including the obstacle detection
module, is shown in the following figure
34
The individual fuzzy controllers utilize the same
membership functions to calculate the collision
possibilities.
Input Variable Distance_From_Obstacle
Output Variable Collision_Possibility
35
  Part of the rules base for left collision.  
 
36

• For left (for right) collision possibilities the
  rule is of the form
• IF A5 is near AND A6 is near AND A7 is near
• THEN collision_possibility is high
• An example of the rules used to extract front
  collision possibilities is
• IF A1 is near AND A2 is near AND A3 is near AND
  A4 is medium_distance AND A5 is near
• THEN collision_possibility is high
• Similarly for the back collision possibility.

37
Second Layer
The second layer fuzzy controller receives as
inputs the four collision possibilities in the
four directions and the angle error, and outputs
the translational velocity, which is responsible
for moving the vehicle backward or forward and
the rotational speed, which is responsible for
the vehicle rotation as shown in Figure.
38
Input Variable Angle Error.
Output Variable Translational_Velocity.
Output Variable Rotational_Velocity.
39

• The angle error represents the difference between
  the robot heading angle and the desired angle the
  robot should have in order to reach its target.
• The angle error takes values ranging from -1800
  to 1800.
• The linguistic variables that represent the angle
  error are Backwards_1, Hard_Left, Left, Ahead,
  Right, Hard_Right, Backwards_2.
• The rotational speed (rad/sec) is described with
  the following linguistic variables right_full,
  right, no_rotation, left, left_full.

40
Example of the rules that control the
vehicle IF Front_Collision is Not_Possible AND
Back_Collision is Not_Possible AND Left_Collision
is Not_Possible AND Right_Collision is
Not_Possible AND Angle Error is Ahead THEN
Translational_Velocity is Front_Full AND
Rotational_Velocity is No_Rotation
41
Test Case 1
Star Desired Target All
the distances are in meters Circle Target
Reached
42
Translational Velocity
Rotational Velocity
______ Left _ _ _ _ Right
Front Collision Possibility
Left and Right Collision Possibilities
43
Test Case 2
Star Desired Target All
the distances are in meters Circle Target
Reached
44
Translational Velocity
Rotational Velocity
______ Left _ _ _ _ Right
______ Left _ _ _ _ Right
Front Collision Possibility
Left and Right Collision Possibilities
45
Test Case 3
Star Desired Target All
the distances are in meters Circle Target
Reached
46
Translational Velocity
Rotational Velocity
______ Left _ _ _ _ Right
______ Left _ _ _ _ Right
Front Collision Possibility
Left and Right Collision Possibilities
47
Results obtained with randomly moving obstacles,
too. Modifications have been made to apply the
same framework to UAV (unmanned aerial vehicles)
abd AUV (autonomous underwater vehicles) Current
activities Fusion of sonars, vision and outdoors
GPS data for testing in outdoor environments,
fleet of cooperating robots.
48
Occupancy Map
Level mapping and binary mapping is used to map
the object into the occupancy map
Back
49
A 2x2 resistor network
A nxn resistor network
Back
50
c
Ç
)
(
M
Area
Where c is the occupancy
ij

c
ij
c
)
(
Area
ij
G is a matrix which contains the conductances of
each node of the network
x
f
)
(
¾
¾

¾
G
C
.
05
3

-



)
.
(
.
exp
.
)
(
x
0
4
2
0
0
10
x
f

-


1


J
V
A
gt
J
A
V
where A is an (n2 ? n2) matrix, called the
admittance matrix V is an (n2?1) matrix
representing the potential values at each node of
the resistor value and J, the current matrix, is
an (n2?1) matrix whose values are non-zero only
at points of application of external current
sources
Back