Robotic Form and Function: Creatures and Humanoids
by Dr. Mark L. Swinson, Acting Director, DARPA Information Technology Office
http://www.roboticsonline.com/public/articles/articlesdetails.cfm?id=496
Reprinted with permission from Sandia National Laboratories, Sandia Technology
magazine, summer 2001. (www.sandia.gov)
Robotic beasts and machines
Many of our robots will have forms dictated by their tasks, environments or
even our imaginations. We can expect to see robotic insects, dogs, and birds
as well as robotic cars, boats, and perhaps washing machines. Robotics may
represent the greatest unmet technological expectation of the twentieth
century, but many of those expectations will likely begin to be realized early
in the twenty-first.
Advances in artificial intelligence technologies are critical to this
progress. One example is situated machine learning. Reinforcement learning can
be used for an unsupervised, learning-with-critic approach where mappings from
precepts to actions are learned inductively through trial and error. Other
approaches may include evolutionary methods that begin with an initial pool of
program elements, and then use genetic operators such as recombination and
mutation to produce successive generations of increasingly successful
controllers.
These approaches (as well as others) will teach robots to adjust parameters,
exploit patterns, evolve rule sets, generate behaviors (and aggregations of
behaviors), devise new strategies, predict changes in the environment, and
even exchange this knowledge with other robots. Such robots can acquire new
knowledge, as well as adapt existing knowledge to new circumstances, and
thereby solve problems in ways we humans may not understand. Indeed, emergent
behavior, rather than being suppressed by careful design, may instead be
encouraged by equally careful design.
As robots become more pervasive, then like automobiles they are likely to
become increasingly complex. Indeed, some robots may be comprised of millions
of parts. If fast, cheap, rapid manufacture of these robots is to occur, it
may be necessary to remove humans from the process altogether. Jordan Pollack
and his colleagues at Brandeis are using commercial CADCAM simulators together
with a genetic algorithm to evolve the body and brains of simple robots. They
have succeeded in automatically designing, improving, and creating a real
robot with only trivial human intervention. So far, the work has focused
solely on creating a robot for locomotion but, eventually, this approach may
allow cheap, near-perfect solutions to be evolved and deployed, even for
complex tasks requiring unintuitive solutions.
Humanoid Robots
Humanoid robotics includes a rich diversity of projects where perception,
processing, and action are embodied in a recognizably anthropomorphic form in
order to emulate some subset of the physical, cognitive, and social dimensions
of the human body and experience, with the goal of creating a new kind of
tool. Such a tool would be intended to work not just for humans but also with
them. Humanoids will be able to work safely along side humans in typical,
everyday environments, as well as the more daunting environments of space and
underseas, and thereby extend our capabilities in ways we cannot at present
imagine.
Indeed, humanoids may prove to be the ideal robot design to interact with
people. After all, humans tend to naturally interact with other human-like
entities; the interface may well be hardwired in our brains. Their
human-like, robot bodies will allow them to seamlessly blend into environments
already designs for humans. While we humans have historically adapted to the
limitations of our machines, here the machines will be designed to adapt to
us. Humanoids will not only provide a new way for us to interact with
machines, but may also serve as an intuitive filter for humans to interact
with an increasingly ubiquitous and pervasive information environment.
Humanoid robots that can incrementally acquire new knowledge from autonomous
interactions with the environment will accomplish tasks by means their
designers did not explicitly implement (or perhaps even conceive of), and will
perhaps thereby be capable of adapting to the unanticipated circumstances of
an unstructured, dynamic environment. Already, humanoid robots have
demonstrated basic task decomposition necessary to carry out complex commands
given through gesture and speech. Humanoids have also demonstrated the
ability to adapt, to orchestrate existing capabilities, and to create new
behaviors using a variety of machine learning techniques.
Humanoid robots may well rekindle a new inspiration for artificial
intelligence as they motivate research toward intelligent, autonomous systems.
Already, a growing number of robotics researchers have found that the human
form provides an excellent platform upon which to enable real-world learning.
A ''similar'' body facilitates learning based upon imitation, by making it
easier to map the human's action onto the robot. In fact, it may be that
human-like intelligence actually requires a human-like body. As a minimum,
the anthropomorphic form factor of these robots enables them to easily
interface with existing technology and infrastructure with minimal disruption.
Hence, humanoids appear to be a uniquely appropriate form factor by which to
gradually introduce 'intelligent' robots into new application domains.
_____________________________________________________________________
«¤»¥«¤»§«¤»¥«¤»§«¤»¥«¤»«¤»¥«¤»§«¤»¥«¤»§«¤»¥«¤»§«¤»¥«¤»§«¤»¥«¤»§«¤»¥«¤
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Stay hungry,
--J. R.
Useless hypotheses, etc.:
consciousness, phlogiston, philosophy, vitalism, mind, free will, qualia,
analog computing, cultural relativism, GAC, Cyc, Eliza, and ego.
Everything that can happen has already happened, not just once,
but an infinite number of times, and will continue to do so forever.
(Everything that can happen = more than anyone can imagine.)
We won't move into a better future until we debunk religiosity, the most
regressive force now operating in society.
This archive was generated by hypermail 2b30 : Fri Oct 12 2001 - 14:40:11 MDT