open in new window

advertisement
Research Trends
in
Artificial Intelligence
Muhammad Younas
Hassan Javid
Danish Shah
05020110
05020304
06020225
Maze – Solving Robot
Robot’s Tasks
 The robot is a simple machine. It can move north,
south, east or west (NSE or W).
 It has no sensory input and does not have conditional
moves; its behavior is determined by a list of which
moves it is to take while in the maze.
 The maze has one entrance (where the robot starts)
and one goal (the robot's goal).
 The closer the robot gets to the goal in the fewest
number of moves, the better its score will be.
Maze – Solving Robot
Robot’s Tasks
 The robot will simply be represented as a list of moves. e.g (n e e s s) etc.
Maze Size can be variable. Such as a 4*4 maze can look like this. E is the
entrance, and R is where the robot current position after executing the
moves given above. Since the robot cannot see anything, it merely moves
until it reaches the goal. How close it gets to the goal determines its score. If
it reaches the goal, it gets a high score and extra pointes are awarded when
it reaches the goal with the fewer steps.
GOAL
WALL
ENTRANCE
WALL
R
Maze – Solving Robot
Definitions
 Genome
The list of a population member's genes. Genes are 1s and 0s. For
example:(1 0 0 1 1 0 1 1)
 Phenome
 Phenome is an expression of an individual's genome in the "world." In this
case, the phenome is the form of what's being evolved, or a description of
the behavior of the individual, and it comes directly from the genome. The
robot's genome would be a list of its moves, decoded from the genome.
 If every 2 bits in the genome encoded for one move (00 = n, 01 = s, 10 = e,
11 = w), then the above genome's phenome would be
GENOME= (1 0 0 1 1 0 1 1)
PHENOME=(e
s
e
w)
Maze – Solving Robot
Definitions
 Fitness
Fitness tells how good a genome is. Genomes are sometimes scored directly, but
often they are converted into a phenome and the phenome is given a score. For
example in GENOME= (1 0 0 1 1 0 1 1) that is PHENOME=( e s e w ) if
the robot's moves of e, s, e, w lead it closer to the exit, it would have a higher score
than one that leads it to some dark corner of the maze far away from the goal.
 Individual
Genetic Algorithms will operate on individuals. An individual will be a list of the fitness,
genome and corresponding phenome. For example
(4 (1 0 0 1 1 0 1 1) (e s e w))
 Population
is group of individuals. The GA operates on these, producing their phenomes from
genomes, evaluating each one's fitness, then selecting mates and producing new
individuals for the next generation. This will be a list of individuals.
Maze – Solving Robot
The Genetic Algorithm
 The Genetic Algorithm will begin with an initial
population with randomly generated genomes.
 It generates phenomes for each genome, then
evaluates each individual's fitness based on its
phenome.
 During each run, we will apply the genetic operations
and delete/create new individuals.
Maze – Solving Robot
The Genetic Algorithm
Some of the routines that we have outlined and implemented
are so far are.
1.
Mutate
Takes a genome, returns the same genome with some of the bits flipped.
2.
Cross Over
Takes 2 genomes, performs one-point crossover on them to produce two new
genomes.
3.
Mate
Takes 2 individuals and performs crossover on their genomes to get 2
new genomes. It then mutates the new genomes. Finally, it makes 2 new
individuals with dummy values for fitness and phenome.
Maze – Solving Robot
The Genetic Algorithm
 Selecting an Individual
Takes a population of individuals. Chooses a single individual randomly and
returns that individual. This random choice is based on the fitness value of a
genome in the population
 Random Population
Takes a population size and number of genes in each genome and
generates a population of individuals with random genomes
 Random Genome
This function takes a genome-length and returns a random genome of that
length.
………………………………………
Maze – Solving Robot
The Genetic Algorithm
 We are almost done with the research for our project
and we have outlines almost complete pseudo code.
 We have started working on implementation and have
implemented some of the routines (functions) that
were listed before.
 Hopefully by next presentation we will come up with
more routines implemented.
References
 Embodied Intelligence at www.ai.mit.edu.
 Genetic Algorithms at genetic algorithm@Everything2.com
 Maze Solving Algorithms at www.aboutAI.net.
 Artificial Intelligence at www.indiana.edu
 John Knight page at www.doe.carleton.ca.
 Tony Pipes page at www.ias.uwe.ac.uk.
Download