Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone...

31
Synthesising Mr. Antony Gerdelan Dr. Napoleon Reyes, Ph.D. Computer Science, Massey University, Auckland, New Zealand Using a Fuzzy Logic, Neural Networks, Evolutionary Algorithms

Transcript of Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone...

Page 1: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Synthesising

Mr. Antony Gerdelan

Dr. Napoleon Reyes, Ph.D.

Computer Science, Massey University, Auckland, New Zealand

Using a

Fuzzy Logic, Neural Networks, Evolutionary Algorithms

Page 2: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

Motivation: Autonomous Navigation Challenges

Central Idea: Integrating Complementary Path-

Finding Systems

Defining Robot Behaviours:

Target Pursuit, Obstacle Avoidance & Opponent

Evasion

Integrating the two complementary systems:

A Novel Hybrid Fuzzy A* Navigation System

Results & Analysis: Robot Soccer Simulations

Conclusions & Future Works

Page 3: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Simulations

3-D Hybrid Fuzzy A* Navigation System

Cascade of Fuzzy Systems

Page 4: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Robot Soccer Set-up

Colour objects

Fluorescent lampsOverhead

Camera

Exploratory environment is indoor – room totally obstructed from sunlight

Multiple monochromatic light sources – fluorescent / fluoride lamps

Colour Object Recognition (Recognition speed: < 33ms)*

Page 5: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

We introduce a new algorithm that is

designed to handle the robot

navigation part of the robot soccer

system and incorporates several

behaviours

Page 6: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

• allot time for vision, team strategy, path-planning

Time Constraints

Finding an Optimal Path in Real-time

• calculate shortest path dynamically

Local minima Problem

• prevent robot from being trapped

Inaccuracy of Sensors

• position, robot orientation, distance from target,

etc. could be inaccurate

Page 7: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

• reactionary algorithm + optimal path planning

• allow for course correction dynamically

• simplify the path-planning algorithm to calculate

only a rough estimation of the route and let the

reactionary algorithm iron-out the complete path

Page 8: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the A* Algorithm

*

S

Page 9: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the A* Algorithm

*

Path-Finding in a Static Environment

• Generally not used in a dynamic environment for

these reasons

• Used in computer games for decades

•‘Tile-based’ environments (fixed nodes)

• Pre-calculation of paths (does not account for

dynamic obstacles)

Page 10: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the A* Algorithm

*

Calculating h* values for a

computerised map:

•Start Node at (1,S1)

•Goal is (13,S13)

•Optimal Path (not

considering obstacles) is

from highest point to

lowest point

Heuristic Map (h*)

Page 11: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the Modified A* Algorithm

*

Evasion Incorporated in

path planning layer:

•Must specify perceived

level of threat to areas of

environment

•A second computerised

map is created. Each

where each cell is

awarded an undesirability

value.

•Function specific to

problem must be created

to award u values. In our

case a teardrop effect

around hostile robots.

Undesirability Map (u*)

Page 12: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the Modified A* Algorithm

*

Undesirability Map (u*)

Undesirability across

environment can also be

represented as a 3D map.

•Z-axis represent

undesirability value of

each cell.

•Higher peaks = more

undesirable areas.

•Similar to potential fields

approach, but also pre-

empts movements and

direction of travel of hostile

agents.

Page 13: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

An Evasive Path-Finding Algorithm

*

The Modified A*. The Figure depicts the sum of (h* + u*)

Combining A* With

Undesirability Maps We

Inculcate an Evasive

Property into our path-

planning behaviour.

A new f* score:

f* = h* + u* + g*

•Balances undesirability u*

with heuristic h*

•Can be represented in 3D for

a hills-and-valleys effect

(pictured).

•Ranges of u* must be very

carefully balanced with h*

values

•g* components ensures

overall steepest downhill

(shortest) path

Page 14: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Complementary Path Finding Systems

*

• Reactive

• Very refined robot movement thanks to C.O.G.

calculation

• Limited long-term path-planning ability

Fuzzy Logic:

A* algorithm:

f* = h* + g*

• Very fast calculation thanks to heuristic h*

• Optimal long-term planning (finds best path if one

exists)

• Paths are very rigid – movement produced is not

natural or refined

Page 15: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Sensors

Environment Information

Nearest

Obstacle and

Final Target

Locations

Environment

Processor

The A* Algorithm

Path Planning Layer

Fuzzy Logic

Reactionary Layer

Map

Actuator Control

Module

Speed and Rotation

Motor Instructions

Next Waypoint

Actuators

Page 16: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the A* Algorithm

*

Specially Modified A* for a Dynamic Environment

Demands of the Problem

• Environment not tile-basedA map (grid) must be created

• Obstacles move dynamically and so the path

can not be pre-calculatedPath must be continually recalculated to take into account moving

obstacles

Our Approach

• Entire path generated, but only the next path

point is passed down to the Fuzzy Logic layerNow, if a moving obstacle disrupts the planned path, it is re-

calculated on the fly

Page 17: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Path-Finding With the A* Algorithm

*

Specially Modified A* for a Dynamic Environment

Because we now only consider the next path

point:

The effective path can ‘cut corners’ in practice

For this reason, Dynamic A* Path Finding is

not effective alone as a navigation system.

Fuzzy Logic is required to smooth out the

effective path

Page 18: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

S

Complementary Path Finding Systems

The Hybrid Fuzzy A* Navigation Algorithm

Page 19: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

S

Disagreements between Layers

Real case scenarios

The A* Algorithm may

direct the robot to

take a straight path

between two close

obstacles, but the

Cascade of Fuzzy

Systems would

engage in obstacle

avoidance.

Page 20: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Complementary Path Finding Systems

*

Resolving disagreements between systems

The Fuzzy Logic layer must be able to navigate the

robot between any two path points generated by the A*

Planning Layer

Tweaking of Fuzzy Set Membership values to solve

some tricky navigation

Page 21: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

S

Strength of the Hybrid Fuzzy A* Algorithm

Real case scenarios

The A* Algorithm

may direct the robot

to take a straight

path crossing the

obstacle, but the

Cascade of Fuzzy

Systems allow for

course corrections.

Obstacles may not be

perfectly sitting on just

one block (or cell). It

may be overlapping

several blocks at any

one time.

Page 22: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Cascade of Fuzzy Systems

Adjusted Speed

Adjusted Angle

Next Waypoint

N

Y

Adjusted Speed

Adjusted Angle

Fuzzy System 1: Target Pursuit

Fuzzy System 2: Speed Control for Target Pursuit

Fuzzy System 3: Obstacle Avoidance

Fuzzy System 4: Speed Control for Obstacle

Avoidance

ObstacleDistance <

MaxDistanceTolerance

and closer than Target

Actuators

Path planning Layer:

The A* Algorithm

Multiple Fuzzy Systems employ the various robot behaviours

Fuzzy System 1

Fuzzy System 2

Fuzzy System 3

Fuzzy System 4

Path Planning Layer

Central

Control

Target

Pursuit

Obstacle

Avoidance

Page 23: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Fuzzy Sets

Angles

Sub ranges for Small, Medium and Large Angles overlap

SMALL

MEDIUM

LARGE

*

Small Medium Large

Deg

ree

of

me

mb

ers

hip

Angle in degrees

Membership Functions

Page 24: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Taking advantage of Angle Symmetry

Fuzzy Sets for Angles

The approach developed takes advantage of the

Angle symmetry, and is equivalent to having 6 Fuzzy

Sets for the Angles.

*

SMALL

MEDIUM

LARGE

SMALL

MEDIUM

LARGE

Left Right

A simplified

approach that

reduces the

computational cost

without sacrificing

accuracy.

Page 25: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Fuzzy Sets

Distances

Sub ranges for Near, Far and Very Far Fuzzy Sets for Distances overlap*

De

gre

e o

f m

em

be

rsh

ip

Distances in cm.

Membership Functions

Near Far Very Far

NEAR

FAR

VERY FAR

Page 26: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Fuzzy Systems for Target Pursuit

NEAR FAR VERY FAR

SMALL Mild Turn Mild Turn Zero Turn

MEDIUM Med Turn Mild Turn Mild Turn

LARGE Sharp Turn Med Turn Med Turn

Next Waypoint

The A* Algorithm

NEAR FAR VERY FAR

SMALL Med Speed Fast Speed Very Fast

MEDIUM Slow Speed Med Speed Fast Speed

LARGE Very Slow Slow Speed Slow Speed

e.g. If the Distance from the Target is NEAR and

the Angle from the Target is SMALL

Then the robot should make a Mild Turn.

Fuzzy System 1 (Steering)

Fuzzy System 2 (Speed Adjustment)

e.g. If the Distance from the Target is NEAR and

the Angle from the Target is SMALL

Then the robot should move at a Medium Speed.

Angle

Speed

Page 27: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

Fuzzy Systems for Obstacle Avoidance

NEAR FAR VERY FAR

SMALL Very Sharp Sharp Turn Med Turn

MEDIUM Sharp Turn Med Turn Mild Turn

LARGE Med Turn Mild Turn Zero Turn

Nearest Obstacle (Distance and Angle)

NEAR FAR VERY FAR

SMALL Very Slow Slow Speed Fast Fast

MEDIUM Slow Speed Fast Speed Very Fast

LARGE Fast Speed Very Fast Top Speed

e.g. If the Distance from the Obstacle is NEAR and

the Angle from the Obstacle is SMALL

Then turn Very Sharp.

Fuzzy System 3 (Steering)

Fuzzy System 4 (Speed Adjustment)

e.g. If the Distance from the Obstacle is NEAR and

the Angle from the Obstacle is SMALL

Then move Very Slowly.

Vision System

Angle

Speed

Page 28: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

ADVANTAGES

*

In the robot soccer game, there are 11 robots to

control at the most.

SPEED

We are able to demonstrate that our algorithm can

control more than 30 robots at twice the required

speed of calculation.

Page 29: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

*

The system is able to take advantage of A*’s

optimized path finding, as well as Fuzzy

Logic’s reactionary feature, allowing for a

significant improvement in speed and

adaptability to a harsh dynamic environment.

We have successfully developed a Hybrid

Fuzzy A* navigation system that inculcates

three adaptive robot behaviours:

• Target Pursuit

• Obstacle Avoidance

• Opponent Evasion

Page 30: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

The algorithm developed is able to combine

the strengths of the two intelligent techniques,

as well as eliminate their weaknesses in one

Hybrid Fuzzy A* architecture

New Predictive quality behaviour has been

inculcated into the system with the addition of

a new parameter in the A* formula.

*

Page 31: Fuzzy Logic, Neural Networks, Evolutionary Algorithms ...nhreyes/MASSEY/159741...not effective alone as a navigation system. Fuzzy Logic is required to smooth out the effective path

By tweaking the range of values assigned to

the undesirability feature we can vary the

degree of aggressiveness of the robots.

Ball Carrier – most evasive (high u* values)

Other Attackers – less evasive (low u* values)

Defender –not evasive (zero u* values)

*