Abstract
Dr. Inman Harvey
"Artificial Evolution: A Continuing SAGA"
I will start with a basic tutorial on Artificial Evolution, and show the simplest possible way of
implementing this with the Microbial Genetic Algorithm. I will then discuss some shortcomings in many
of the basic assumptions of the orthdox Genetic Algorithm (GA) community, and give a rather different
perspective. The basic principles of SAGA (Species Adaptation GAs) will be outlined, and the concept
of Neutral Networks, pathways of level fitness through a fitness landscape will be introduced. A
practical example will demonstrate the relevance of this.
Prof. Robert Full
"Using Biological Inspiration to Build Artificial Life that Locomotes"
Nature's processes and designs can assist us in the construction of life-like robots.
1. Nature's general principles can provide biological inspiration for robotic designs. The blind
copying of nature is likely to fail, since evolution works on the "just good enough" principle.
Organisms not only bring to us amazingly adaptive designs, but also carry with them the baggage of
their history. Nature co-opts the parts it has for new functions. Adult parts often serve multiple
functions and have been constrained by development. Nature provides useful hints of what is
possible and design ideas that may have escaped our consideration. These biological principles
should be explored in our best engineering efforts.
2. The discovery of general biological principles requires a collapse of dimensions in complex
systems. Behavior results from complex, high dimensional, nonlinear, dynamically coupled
interactions of an organism with its environment. Animals show kinematic, actuator and neuronal
redundancy when a single behavior such as locomotion is examined. Reducing redundancies
by seeking synergies yields simple, general principles. For instance, animals that differ in leg number,
body form and skeletal type show the same motions - legged animals bounce like people on pogo
sticks. Force patterns produced by the six-legged insects are the same as those produced by
trotting eight-legged crabs, four-legged dogs and running humans.
3. Nature will become an increasingly more useful teacher as human technology takes on more
of the characteristics of nature. Even if we had all the general biological principles, we don't
have the technology to use them effectively. Information handling has changed dramatically, but
until recently the final effectors (metal beams and electric motors) have not. Traditionally, human
technologies have been large, flat, right-angled, stiff, and rotating, with few actuators and sensors,
whereas nature is small, curved, compliant using appendages with multiple actuators and sensors.
Human technology is changing with the greater use of nonmetallic, more flexible materials and
increased miniaturization. Revolutionary new technologies in materials and manufacturing promises
to lead to more life-like, mobile robots in the future when inspired by nature.
Dr. Hiroaki Kitano
"Integrated Perception and Behaviors in Humanoid"
This talk presents perception and behavioral control in a humanoid.
One of the essential aspects in intelligent systems research is how to
attain robustness in perception and behaviors. Much can be learned from
biological systems that are living proof of robust systems.
Our research focuses on robust systems by integrating multimodel perception
as well as adaptive behavioral control, particularly for bi-ped walking.
Prof. Jordan Pollack
"Coevolutionary Robots"
I address the problem of robotics, from economic, computational, and biologically inspired
perspectives.
Economically, robots have been a failure. Stimulated by fantasy literature, humanity has been expecting
humanoid robots to become our ethical slaves, but so far we've only gotten ATM machines and ink-jet
printers. Expensive to develop and manufacture, complex robots are just bodies without brains, electronic
puppets controlled by humans with joysticks or scripts, waiting for the AI problem to be solved. Few
robots ever provide a return on investment.
In nature, the bodies and brains of creatures arise together, the result of a long series of small mutual
adaptations. There is never a situation in which the hardware has no software, or where a growth or
mutation - beyond the adaptive ability of a brain - survives. This chicken-egg problem of body-brain
development may be best understood as a form of co-evolution.
In our lab we have been studying coevolutionary machine learning as an approach to self-organization. The
basic idea is that as learner adapts to an environment, the environment adapts to the learner, creating
new challenges which (theoretically) lead to a local anti-entropic spiral of complexity. We have had many
limited successes in games, optimization, language, and problem solving, before turning to the robotics
problem.
Using evolutionary computation, neural networks, physical simulation, and constraints from manufacture,
our software generates blueprints and controllers for robots automatically. These are then fabricated from
reusable sensors, effectors, and chips, held together with modular parts (like Lego) or by rapid
prototyping (robotically manufactured).
Besides pointing the way to really cheap disposable robots, our work also illustrates how design can
emerge without a designer from the interaction of evolution and physics.
Dr. Owen Holland
"From the Imitation of Life to Machine Consciousness"
In 1949, Grey Walter, an English neurophysiologist, produced the first autonomous mobile robot designed
to be 'an imitation of life'. He identified a number of characteristics typical of living beings, and
showed that his robot, the famous 'turtle' or 'tortoise', could demonstrate analogues of these
characteristics in normal room environments. Remarkably, the robot achieved all this using only two
vacuum tubes, or 'brain cells'. How was this possible? Recently discovered documents show that Grey Walter
used an early form of the behaviour-based technology that is now dominant in the design of small robots.
The first part of this talk details and analyses Grey Walter's achievements, emphasising that many of his
insights and methods are still relevant today. Although he left little in the way of formal records of
his research, it has been possible to reconstruct much of what he did from contemporary photographs and
films, and the talk draws heavily on these sources.
The second part of the talk examines a modern enterprise as challenging as Grey Walter's search for
'an imitation of life': the quest for machine consciousness. If successful, this will have profound and
far-reaching effects on science, on society, and on industry. How can we approach this problem? Do we
yet know enough about consciousness to even begin? Is the necessary technology available? How will we
know if we have succeeded? And what part will robots play in all this? These and other questions are
discussed, and several examples of current research into the subject are presented. In 1994, Francis
Crick hoped that: "...before long, every laboratory working in the visual system of man and the other
vertebrates will have a large sign posted on its walls, reading CONSCIOUSNESS NOW". The talk argues that,
before long, this sign will be seen in many robotics laboratories too.
Dr. Dario Floreano
"Evolution of Spiking Neural Controllers for Autonomous Vision-based Robots"
Vision is a major source of sensory information for robots that live and operate in our natural
environments. Current electronics technologies, such as CMOS and analog VLSI circuits, allow the
construction of miniature, dense, and low-power vision circuits that can be embedded in a
variety of autonomous robots, ranging from 1 inch mobile robots to micro air vehicles. Given the
constraints of energy consumption and real-time operation it is unlikely that general-purpose,
number-crunching image processing techniques are suited for these robots. In this talk we shall
describe two sets of experiments where we take an evolutionary approach to investigate simple, fast,
and energy-efficient mechanisms to perform complex vision-based behaviors. In both cases, the idea is
to exploit dynamical properties of circuits made of simple components.In the first set of experiments,
we evolve circuits of spiking neurons for robust vision-based navigation. These networks, which can be
implemented in low-power analog VLSI circuits, are difficult to design for a given functionality, but
are easy to evolve and outperform similar networks of conventional McCulloch-Pitts neurons for their
rich dynamical behaviors. Instead, in the second set of experiments we assume that we have a
micro-processor with limited computational abilities and evolve an active vision system that is free to
scan the visual field in order to discover simple and relevant features that are useful for the visual
task. We show a few examples where such an evolved vision system can discriminate between complex shapes
where multi-layer neural networks trained with supervised algorithms fail. We also show how the system
can be used for a task of robot navigation with a mobile camera.
Prof. Rodney Brooks
"Steps Towards Living Machines"
We still do not understand what it is that makes living matter alive. If we did we could build living
machines, but it is clear that we do not have the technology to do that today.
Living machines would be able to self-reproduce, find their own sources of energy, and repair themselves
to some degree. They need not necessarily be built from our standard materials, silicon and metal.
Living machines will change all of our technologies with equivalent disruption as that introduced by
electricity and that by plastics. Living machines will invade the fabric of our everyday lives.
There are three thrusts to trying to build living machines. First is to build robots with partial
characteristics of living machines, looking for the key intellectual ideas that make them possible.
The second is to use generalized evolutionary systems to investigate possible mechanisms and designs.
Generalized evolutionary systems use analogs of physical processes to organize the world for evolving
systems, living in that world. The third thrust is to develop a new mathematics of living systems. This
new mathematics interacts with the first two thrusts in two ways. It is inspired by the first two thrusts
to formalize the notions developed there. Additionally it is used to provide constraints on the design
spaces in the first two thrusts, to guide the research work to the appropriate areas.
Dr. Francesco Mondada
"Interactions between Art and Mobile Robotic System Engineering"
Mobile robotics offers a new media for public entertainment and art. Mobile robots can display behaviour
in the real world and this behaviour can be a new artistic or entertainment support, very different from
sculptures, drawings or video. This new artistic or entertainment support, like other technological
supports like video or Internet, need a strong and very specific technical know-how. Most of the actual
robotic art and even entertainment does not exploit the potentiality of the mobile robotics field. Many
artists exploit only the motion aspect, but are unable to create real behaviours. Others exploit the
effect of the existing robot image with simplistic realisations. Only few artists try to really integrate
mobile robotics in their approach. This integration is an important effort, mobile robotics being itself
an interdisciplinary field including artificial intelligence, electronics, computer science, mechanics
and much more. The design of a mobile robot need a strong interdisciplinary and system oriented
engineering process. The addition of artistic or entertainment constrains bring a new dimension to the
problem and need a strong coherent approach.
This paper illustrates this interdisciplinary approach with six examples of projects of robotic art or
entertainment made during my work at K-Team in the last 5 years. Some of them are very simple
realisations, others have received prizes in international artistic exhibitions. All have similar
problems and need similar methodologies. Several issues are discussed, including the effects on
visitors (positive or negative, objective or subjective, intentional or not), the problems of interactions
between artists and engineers, aspects of mechanic, electronic and behavioural design applied to
entertainment and other issues developed during specific projects. These examples will be used to
illustrate methodological issues needed in this interdisciplinary work.
|
© 2008, AAI Canada, Inc.
112 Cavanaugh Drive, Ottawa,
Ontario, Canada K0A 1L0
TEL: +1.613.839.6161 or 1.800.895.1122 ; FAX: +1.613.839.6616
AAI Canada, Inc. provides this information as an informational tool to the public,
and is not responsible for any errors in content. All information should be verified by the reader.
General Inquiries: Contact Us
Technical Inquiries: Webmaster
|
|
