An introduction to Autonomous mobile robots

An introduction to
Autonomous
mobile robots
Zahra Sadeghi

2
introduction
• autonomous robot :
a machine that operates in a partially unknown and unpredictable
environment.
– cannot always be programmed to execute predefined
actions .
• one does not know in advance what will be the universe of required
sensorimotor transformations required by the various situations that
the robot might encounter.
– the environment might have dynamic characteristics that
require rapid online modifications in the robot behaviour.
in the last ten years several researchers have looked at novel
methods for setting up autonomous mobile robots.

3
Why simulation studies?
• researchers working in the field of artificial intelligence
– software engineers or academics coming from various
disciplines like biology, psychology, or anatomy.
– These people are not very keen to deal with mechanical and
electronical problems.
• Only few laboratories in the world have the possibility to
build a robot and program it
– Need very specific programming languages and thus were not
accessible to outsiders.
– Making a real robot seems often too expensive in comparison to
simulations.

4
Khepera miniature mobile robot
• Lack of simple, efficient, robust and low cost robotic
tools.
• intended as a tool that could be easily used by a
variety of people with different back grounds and
needs,
– such as biologists willing to test theories of adaptation,
– neurophysiologists developing models of neural networks,
– psychologists implementing models of learning and
cognition,
– and engineers evaluating different control strategies for
specific applications.

5
khepera
• viable compromise:
– 55 mm in diameter and a variable height
– sufficiently small to easily move on a desk and sufficiently big to be built
with standard electronic and mechanical components.
• advantages of its size:
I. complex environments on a limited surface
– a normal office desk of 0.9m x 1.8 m
– one can use a thin suspended cable for power supply and also place a camera
with normal lenses on the ceiling for monitoring the whole environment.
II. Fundamental laws of physics give higher mechanical robustness to a robot
of this size.
– compare a robot of 50 mm in diameter crashing against a wall at 50 mm/s with a
robot of 1 m in diameter crashing against the same wall at 1 m/s.
The miniature robot will resist the collision, the other robot will probably report
serious damages.
III.The price is lower than that of a robot with comparable performance, but
larger size.
– due to smaller parts smaller circuits,
– smaller infrastructure to build and test the robots,
– simpler mechanics

6
HARDWARE MODULARITY
Khepera is based on this concept of modularity, both in hardware and software.
• Advantages:
– possible extensions :one can employ additional computational devices
on extension modules, thus keeping the main processor free for global
management of the robot behavior.
– cheaper equipment :enables different possible configurations and
experiments using the same basic components.
• Architecture:
– has an extension bus that makes it possible to add turrets on the top of
the basic configuration, depending on the needs of the experiments to be
done.
– This modularity is based on a parallel and a serial bus.
• parallel bus can be used for simple extensions directly under control
of the main Khepera processor.
• serial bus implements a local network for inter-processor
communication. Using this second bus, other processors can be
connected to the main one in order to build a multiprocessor structure
centered on the Khepera main processor.

7
Software modularity
• Advantages:
– flexibility and possibilities for extensions,
• Architecture:
– consists of a flexible protocol that recognizes all added
extension modules when the robot is powered, informing
the main processor about all functionalities available in
each extension as well as the procedures for activating
these functionalities.
– The BIOS, is also based on a modular structure.
– Motors, sensors, and timing functions are grouped into
distinct modules to simplify management of the robot and
improve software robustness.
– The main software also supports remote control and
downloading of specific applications through a serial
cable.

8
KHEPERA CHARACTERISTICS
• In its basic configuration, Khepera consists of two layers:
– the sensory-motor board and the CPU board.

9
• The motor system consists of two lateral wheels and two pivots on the front and
back.
• The sensory system is placed on the lower board, consisting of 8 infrared-light
proximity sensors distributed around the body, 6 on one side and two on the other
(this asymmetry can be used to establish the front and back of the robot).
– These sensors can detect the presence of objects by emitting and measuring
reflected light and
• sensorimotor board
– Are also placed NiCd batteries with a capacity of 110 mAh which allow
the robot to be self-sufficient for approximately 30-40 minutes.
• CPU board
– encloses the robot's main processor
a Motorola MC68331 with 128 K-bytes of EEPROM
and 256 K-bytes of static RAM
• A/D converter allows the acquisition of analog signals coming from the
sensory-motor board.
• RS232 serial line :a wire can also provide continuous power supply from an
external source.

10
extensions
• Possible extensions :
• The gripper module
– is connected on the extension bus
– has its own processor which communicates with the main
Khepera processor using the local network.
– The gripper has two degrees of freedom:
• elevation and grasping.
• Sensors inform about elevation of the arm, position of the gripper fingers,
presence of an object inside the gripper (optical barrier), and electrical
resistivity of the grasped object.
• The K213 linear vision
– can be placed on the top of the gripper
– has its own local processor which uses the local
network to share information with the main processor.
– has a horizontal linear camera composed of
64 photo-receptors giving a gray-level image
– and an auto-iris function for automatic adaptation
of image sensitivity to changing light conditions which also provides
information on global light intensity.

12
Fuzzy control
• Fuzzy logic offers the possibility to express and implement human know-how
in the form of linguistic if-then rules which can be applied for the control of
nonlinear systems, such as mobile robots
• Every rule has two parts:
– the antecedent part (premise), If. . . ,
description of the state of the system which is used to activate one rule,
– the consequent part, : then. . . .
action that the operator who controls the system must take.
• The general form of a linguistic if-then rule is:
If a set of conditions is satisfied
then a set of consequences can be inferred.
• linguistic variables :The process states and control variables.

13
A fuzzy controller is composed of four principal modules:
• fuzzification interface: crisp values fuzzy sets.
• knowledge base: supplies the fuzzication module,the inference engine, and
the defuzzication interface with necessary information for their proper
functioning.
• decision making unit (or inference engine) : computes the meaning of the set
of linguistic rules.
• defuzzication interface: union of fuzzy sets crisp output.

14
NEURO-FUZZY CONTROL
• an automatic method for the design and the analysis of a fuzzy
controller
(for two different behaviors: obstacle avoidance and wall-following)
• this method is based on a supervised learning procedure developed
in the frame of neural networks
• design of a neuro-fuzzy controller:
– The first step : to establish a set of initial rules and parameters of the
membership functions.
• During the system operation, a supervisor guides the robot through the
environment while the parameters of the membership functions and of the
rules are adjusted by a learning algorithm
– The second step : the implementation of the controller.
• After learning, the parameters define a new set of rules which can be used
for autonomous operation of the robot. Additionally, the learned rules can be
extracted in a form intelligible for a human being.

15
• After 2000 iteration after 3000 iteration
• using on-line learning instead of memorizing learning vectors
which would take large memory and computational resources.
• Seldom encountered situations need long time to be learned,
• An interesting feature of adaptive neuro-fuzzy control :
learning can start from a set of rules that the engineer might think are
important (instead of starting from random parameters)
and modify them as required by the environment.

16
Evolutionary Robotic
• is a technique for automatic creation of control
systems for autonomous robots
– Genes :the basic “instructions” for building an organism
– Chromosome : a sequence of genes
• A population of different artificial chromosomes, each encoding the
instructions to build a different neurocontroller, is decoded and tested on
the robots.
• While each robot freely interacts with the environment according to the
decoded control system, a fitness function automatically assesses its
performance.
• New populations of increasingly better individuals are created by
repetitively applying selective reproduction, crossover, and mutation to the
evaluated chromosomes for several generations

17
Introduction to Genetic programming
• One of the central challenges of computer science :
to get a computer to solve a problem without explicitly
programming it.
– It is a biologically inspired technique
– automatically creates computer programs to solve problems
– using the Darwinian principle of natural selection, recombination (crossover),
mutation, gene duplication, gene deletion,
• Selection replicates the most successful solutions found in a population at
a rate proportional to their relative quality
• Recombination decomposes two distinct solutions and then randomly
mixes their parts to form novel solutions
• Mutation randomly perturbs a candidate solution
• Why Genetic Algorithms (GAs) ?
– Many real life problems cannot be solved in polynomial amount of time
using deterministic algorithm
– Sometimes near optimal solutions that can be generated quickly are
more desirable than optimal solutions which require huge
amount of time

18
Fitness function
Darwin’s “survival of the fittest”
• A Fitness function evaluates individuals;
• highly fit individuals represent better solutions to the objective
function
Given: desired output for each of a range of inputs
Run: the program on each input,
measure error against desired output
Calculate: an average (or similar) of the error fitness.
• The role of the engineer :to design a fitness function
that measures the performance of the robots with respect to the
desired behaviour or task.

19
The basic genetic algorithm
• Start with a large “population” of
randomly generated “attempted solutions” to a problem
• Repeatedly do the following:
– Evaluate each of the attempted solutions
– Keep a subset of these solutions (the “best” ones)
– Use these solutions to generate a new population
• Quit when you have a satisfactory solution (or you run out of
time)

20
A really simple example
• Suppose your “organisms” are 32-bit computer words
• You want a string in which all the bits are ones
• Here’s how you can do it:
– Create 100 randomly generated computer words
– Repeatedly do the following:
• Count the 1 bits in each word (evaluate)
• Exit if any of the words have all 32 bits set to 1 (quit if satisfactory)
• Keep the ten words that have the most 1s (discard the rest) (best subset)
• From each word, generate 9 new words as follows: (generate new..)
– Pick a random bit in the word and toggle (change) it
• Note :this procedure does not guarantee that the next “generation” will have more
1 bits, but it’s likely

21
Another solution
– Create 100 randomly generated computer words
– Repeatedly do the following:
• Count the 1 bits in each word
• Exit if any of the words have all 32 bits set to 1
• Keep the ten words that have the most 1s (discard the rest)
• From each word, generate 9 new words as follows:
– Choose one of the other words
– Take the first half of this word and combine it with the second half of the other
word
• Half from one, half from the other:
– 0110 1001 0100 1110 1010 1101 1011 0101
1101 0100 0101 1010 1011 0100 1010 0101
0110 1001 0100 1110 1011 0100 1010 0101
• Or we might choose “genes” (bits) randomly:
– 0110 1001 0100 1110 1010 1101 1011 0101
1101 0100 0101 1010 1011 0100 1010 0101
0100 0101 0100 1010 1010 1100 1011 0101
• Or we might consider a “gene” to be a larger unit:
– 0110 1001 0100 1110 1010 1101 1011 0101
1101 0100 0101 1010 1011 0100 1010 0101
1101 1001 0101 1010 1010 1101 1010 0101

22
evolution of straight navigation and
obstacle avoidance
• The fitness function is based on three variables
which are measured at each time step using sensor and motor activations:
– the first one is maximized by speed,
– the second by straight direction,
– and the third by obstacle avoidance.
• All the 80 chromosomes of the initial population are randomly
generated
• each individual is separately tested in the environment.
• After approximately 45 generations, an individual is born which
exhibits smooth navigation around the maze

23
COMPLEX ENVIRONMENTS GENERATE COMPLEX
CONTROLLERS
• Environment:
– a 40x45 cm arena delimited by walls of light-blue
polystyrene.
– A 25 cm high tower equipped
with 15 small DC lamps
is placed in one corner and
the room does not have other light sources.
– Under the light tower, a black-painted
sector is intended to simulate the platform
of a battery charger under construction.
• When the robot happens to be over the black area, its
simulated battery is instantaneously recharged.

24
• Robot:
– Beside the eight infrared sensors, the neurocontroller
receives input also from two ambient light sensors,
each on one side of the body.
– Additionally, another ambient light sensor is placed
under the robot platform, pointing downward,
to detect when the robot arrives on the black-painted sector.
– The robot is provided with a simulated battery
characterised by a fast linear discharge rate
(max duration: approx. 20 seconds),
and with a simulated sensor giving information about the battery status.
• The fittness function is accumulated at every time steps.
• robots that could learn to discover the position of the battery charger and
periodically return to it would survive longer.
• After 240 generations of continuous fittness progress without human in-
tervention, a robot is born which could navigate in the environment and
periodically return to the charging station.

25
CO-EVOLUTIONARY ROBOTICS
• to make the environment more complex and dynamic,
• In the simplest scenario:
– of two co-evolving and competing populations
(for example, predator and prey),
– fitness progress of one species is achieved at
disadvantage of the other population's fittness.
• A competition ends either when the predator
touches the prey or after 500 sensorimotor
updates

26
• Right: the predator is equipped with vision module
• Left: the pray has a black protuberance which can be
detected by the predator everywhere in the environment.

27
• 1. the predator tends to stop
in front of walls,while the prey
moves in circles
• 2.the prey moves fast at straight
Trajectories avoiding walls
while the predator tracks it from
the centre and quickly
attacks when the prey is closer

28
• 3.the predator intercepts the prey which
accelerates causing the predator to
crash against the wall.
Around generation 75, we have
a typical example of the best prey
• 4.pray moves in circles and, when the
predator gets closer, it rapidly avoids it.

29
• 5.As soon as the prey begin again moving
around the environment, the predator
develops a “spider strategy“:
it slowly backs until it finds a wall where it waits
for the fast-approaching prey
• However, this strategy does not pay off when
the prey stay in the same place
at generation 99 we have a new
interesting strategy
• 6.the predator quickly tracks and reaches
the prey which quietly rotates in small circles.
• As soon as the prey senses the predator, it backs
and then approaches the predator (without touching it)
on the side where it cannot be seen;
• consequently,the predator quickly turns in the attempt to visualize
the prey which rotates around it.

30
• In evolutionary or learning single-robot experiments:
the robot is connected to a workstation through a cable which
supplies required energy and supports intensive computing.
• However, it is impossible to use such connections for groups of
more than two robots because the cables would become
entangled.
SPECIFIC HARDWARE TOOLS

31
HARDWARE TOOLS
• a special floor board
as an interface between an external energy supply source and the robots.
a communication link between workstation and robots
• a hierarchical communication strategy
– robot-to-robot (local path, via infrared link)
• Communication among robots is limited to neighbouring teammates.
– workstation-to-robot (global path, via radio link)
• communication enables supervision of the adaptive process

32
Applications of Autonomous
Robots
• Despite the potentials of the approaches and algorithms industrial
applications of autonomous mobile robots are not yet
widespread.
• Neural networks, genetic algorithms, and neuro-fuzzy approaches
– do not offer yet enough guarantee of stability in
unsupervised situations for being accepted as a viable
method in critical applications where risks and costs rank
high.
• Customers
– will read a short set of recommendations, power the robot, and check
from time to time that the work is being properly done.
– are not willing to spend time in instructing their robot, letting it carefully
explore the environment, and buy the risk of sub-optimal performance.

33
FULLY AUTONOMOUS APPLICATIONS
• Husqvarna lawn-mower for flat and prepared terrains.
– 15 cm high .
– has a surface of 80 cm by 40 cm
– covered by solar power cells which let it work for several months when the sun is
high over the sky (a small battery back-up is used by the processor when sun
light is not strong enough to power the robot).
– moves randomly,
– In case the robot gets stuck in unexpected situations, it starts beeping and
waiting for human help.
• Pool cleaning robots
– several types are available on he market
(e.g., see http://h2o-marketing.com/aquabot/aqua.html)
– perform a random walk on the bottom and on the sides while scrubbing, vacuum
cleaning, and filtering the pool.
– They are generally powered via a cable hanging from the centre of the pool and
can also be remotely controlled, if necessary.
• use when the environment is simple and stable enough to expect reliable
operation

34
PARTIALLY AUTONOMOUS APPLICATIONS
• The Mars Sojourner
(http://mpfwww.jpl.nasa.gov/default.html)
– limited be-havioural autonomy.
– Since it moves very slowly, a rather small area of solar cells
is sufficient to power the robot.
– It receives instructions from Earth on its destination
• autonomous wheelchair
– Several handicapped persons found it difficult to steer
precisely their own wheelchair to get around corners or
passing through doorways.
– By supplying these chairs with additional sensors and
appropriate control systems that support semi-autonomous
navigation, the owner can instruct the chair on the desired
destination and let it get there autonomously.

An introduction to Autonomous mobile robots

More Related Content

What's hot

Viewers also liked

Similar to An introduction to Autonomous mobile robots

More from Zahra Sadeghi

Recently uploaded

In this document

An introduction to Autonomous mobile robots