| |
|
|
|
|
|
|
ABSTRACT
This document is a summary of a proposal produced in October
2003, inspired by the visionary FP6 objective
"To construct physically instantiated ... systems that can
perceive, understand ... and interact with their environment,
and evolve in order to achieve human-like performance in
activities requiring context-(situation and task) specific
knowledge"
We assume that this
is far beyond the current state of the art and will remain so for many
years. However we have devised a set of intermediate targets based on
that vision. Achieving these targets will provide a launch pad for
further work towards the long term vision.
In particular we aim to advance the science of cognitive systems
through a multi-disciplinary investigation of requirements,
design options and trade-offs for human-like, autonomous,
integrated, physical (e.g. robot) systems, including requirements for
architectures, for forms of representation, for perceptual mechanisms,
for learning, planning, reasoning, motivation, action, and
communication. The results of the investigation will provide the basis
for a succession of increasingly ambitious working robot systems
to test and
demonstrate the ideas. Devising demanding but achievable test scenarios,
including scenarios in which a machine not only performs some
task but shows that it understands what it has done, and why, is
one of the challenges to be addressed in the project. Preliminary
scenarios have been proposed. Further scenarios, designs and
implementations will be developed on the basis of (a) their potential
contribution to the long term vision, (b) their achievability and (c)
the possibility of practical applications.
Tools will be developed to support this exploration. The work
will use an `open' framework facilitating collaboration with a variety
of international projects with related objectives.
The problem
Despite impressive progress in many specific sub-topics in AI and
Cognitive Science, work on building integrated cognitive systems
moves slowly. Most systems able
to perform complex tasks that humans and other animals can perform
easily, for instance robot manipulators, or intelligent advisers, have
to be very carefully crafted, normally their field of expertise is very
narrow, and they are hard to extend. Whatever intelligence they have
could be described as `insect-like', with very little flexibility or
self-understanding. Part of the reason for this is that over the last
few decades research has become highly fragmented: with many individuals
and research teams focusing their efforts on narrowly defined problems,
for instance in vision, or learning, or language processing, or problem
solving, or mobile robotics.
We propose to try to overcome these limitations by using ideas from
several relevant disciplines to investigate an ambitious vision of a
highly competent robot, combining many different capabilities in a
coherent manner, for instance a non-trivial subset of the capabilities
of a typical human child a few years old. This work will pursue two main
types of objectives concerned with theory and
implementation, and related subsidiary objectives.
Theory objectives
We aim to produce a body of theory, at different levels of abstraction,
regarding
requirements, architectures, forms of representation, kinds of
ontologies, types of reasoning, kinds of knowledge, and varieties of
mechanisms relevant to embodied, integrated, multi-functional
intelligent systems. The results should be useful both for enhancing
scientific understanding of naturally occurring intelligent systems
(e.g. humans) and for the design of artificial
intelligent systems.
We expect such a theory to be built around the core idea of a
self-modifying architecture combining many capabilities which develop
over time, which are deployed concurrently and which interact with one
another asynchronously. The theory would cover both analysis of
requirements for such an architecture and also design options
with their trade-offs. Sub-theories would be concerned with different
sorts of components of the architecture and the forms of representation
and varieties of knowledge that they can use. Theory construction will
build on empirical results in related disciplines (e.g. psychology and
linguistics) and achievements of computer science, software
engineering and AI (including robotics).
Since different sorts of designs are possible the theory will include an
analysis of architectural options and trade-offs as well as
design-options and trade-offs concerning components.
Implementation objectives
We expect to produce well-documented implementations of a succession of
increasingly sophisticated working systems demonstrating applications of
parts of the theory, e.g. in a robot capable of performing a diverse
collection of tasks in a variety of challenging scenarios, including
various combinations of visual and other forms of perception, learning,
reasoning, communication and goal formation. Initially two main kinds of
robot will be investigated both of which will learn from and interact
with human teachers. One of them, the Explorer, will be concerned with
finding its way around a complex building, showing others where to go
and answering questions about routes and locations. The other robot, the
PlayMate, will be concerned with manipulation of structured objects on a
table top. These two require somewhat different physical, perceptual,
reasoning, planning and learning capabilities, though both can be
combined with natural language capabilities and some social competence.
A parallel strand of investigation (the Philosopher scenario) will
investigate common requirements for self-understanding, for reasoning,
for motivation, and for reflection on what is perceived and learnt.
A common problem in all these tasks is deciding where to draw the
nature/nurture boundary, between what is designed in initially and what
arises through learning and development. Alternative options will be
explored. Another common feature is the requirement to integrate
different forms of representation, needed for different
sub-capabilities. Another is the requirement for the robot to have
multiple ontologies which are used as appropriate for different
activities. We shall avoid dogmatism on all these issues, exploring
various alternatives and analysing trade-offs.
Subsidiary activities
The project will also produce a succession of workshops and summer
schools, publications, and an `open' web site containing code,
development tools, theoretical papers, various kinds of re-usable
libraries, demonstration packages, etc, including contributions from
external collaborators, academic and industrial. We expect to have to
share development of tools with other projects.
Print this page
|
|
|
