Spatial Reasoning and Perception in a
Humanoid Robot
Contents:
The Spatial Reasoning and Perception in a Humanoid Robot
Project continues our earlier
work on the Cognitive Robotics I and Cognitive
Robotics II projects. You can read the
final report of Cognitive Robotics II here
(pdf); and can view a presentation (April 2004) about our work here.
The term "cognitive robotics" was
first introduced by the late Ray Reiter who lead a research group
on this topic at the University of Toronto. Cognitive
Robotics intends to capture the application of well founded logical
formalisms and computational models of high-level cognitive
functions, such as perception and planning, and apply this to
robotics, both to real-world and simulated robots.
The approach of the Imperial College Robotics Group differs
in emphasis from that of most research in the Cognitive
Robotics community. In particular, other approaches typically
draw a sharp line between high-level reasoning and the
low-level processing of sensor data, and thereby treat
perception as a black box for delivering facts about the
environment directly to the robot's deliberative component. By
contrast, the Imperial College group takes an integrated
approach to perception, action and cognition, in which
perception is as much open to a logical treatment as, say,
planning or reasoning about other agents.
The Imperial College approach is characterised by the use
of abductive reasoning (reasoning from observations to
possible causes) to provide explanations about current sensor
input in the context of a current description of the world and
a background theory defining (among other things) the
interactions between the robot's environment and its sensors.
Equally notions of change and temporal sequence are
encapsulated in the Event Calculus, a well understood and
mathematically rigorous formalism for reasoning about action,
change, and space (as described in Murray Shanahan's book
"Solving
the Frame Problem: A Mathematical Investigation of the Common
Sense Law of Inertia").
The overall motivation for this project is to use an
upper-torso humanoid robot, LUDWIG, with its particular set of
sensory-motor capabilities, as a test bed for developing
foundational ideas in knowledge representation and reasoning.
The Cognitive Robotics group at Imperial College will be
hosting the TAROS-05
robotics conference in September 2005.
The general aim of our Cognitive Robotics programme of
research is to design and build software for controlling real
(as opposed to merely simulated) robots which (i) are based on
formal logic and the Event Calculus in particular, but which
(ii) take account of the lessons learned in twenty five years
of Robotics practice, in particular the advantages of tight
coupling between sensors and actuators. We combine an Abductive, Event Calculus based
approach with a framework for qualitative spatial reasoning to
create a formal framework for reasoning about perception, space, shape, action,
and continuous change. Using this formalism, theories are
constructed which describe the robot's interaction with its
world.
Spatial Reasoning and Perception in a Humanoid Robot
is a two year research project, which started in December 2002. The project is a
continuation of previous work on the EPSRC funded projects "Logic for Commonsense Reasoning
about Continuous Change", "Cognitive
Robotics I" and "Cognitive
Robotics II". Cognitive Robotics I involved
a map-building task for a single mobile robot and a logic programming
approach to representation and implementation was adopted. The
project also developed logic-based control techniques which
can exploit abductively constructed maps for planning and
navigation. In Cognitive Robotics II this was extended to
cover richer visual sensor modalities, such as stereo-vision
and the use of richer logical descriptions of space and the
explicit modelling of various monocular psychophysical depth cues such
as spatial occlusion or interposition.
We identified several areas leading on from our previous
project work which merited additional detailed research:
- Continued development and exploitation of viewpoint-based logical formalisms for reasoning about
three-dimensional space using three and two-dimensional
visual sensor data. One of the main contributions of
the previous project was the development of a region-based theory of visual space that
allowed arbitrary-shaped bodies to be modelled and
reasoned about in the presence of spatial occlusion
events. This mapped bodies as three-dimensional
volumes of space and viewpoints, to their corresponding
two-dimensional images; facilitating the generation of
hypothesized bodies from images via the application of
abductive reasoning.
- The importance of augmenting earlier formal accounts of
abductive reasoning to to task of active perception and
image interpretation. In the previous project an
earlier abductive account of sensor data interpretation
was generalised to allow for feedback and
expectation. In this project the framework is
extended further to interpret sensori-motor actions and top-down
driven hypothetico-deductive reasoning. Here the
humanoid-robot interacts with objects in its local
workspace and through targeted actions and interaction
with objects, better object representations are
subsequently built up.
The Spatial Reasoning and Perception in a Humanoid Robot
project has already given rise to several significant
results. These include a generalised abductive model of
visual perception, that in one example, maps the three-dimensional object shape
of an object to a finite set of two-dimensional images of the pose (or
'aspect') of objects represented by an Aspect Graph.
By tracking the object using selected features, and
interpreting the change of pose in response to some nudging
action by the robot, potential ambiguity of the interpretation of an object from a set
of aspects is thereby reduced. Other related work has shown that
the known representational weakness of several region-based (or
one-dimensional interval-based) formalisms for modelling
real-world objects, can be jointly combined and implemented in a relational-based
region tracking algorithm. Here conceptual neighbourhood
diagrams (that directly encode direct transitions between
pairs of defined jointly exhaustive and pairwise disjoint n-ary
relations and which have been used to measure similarity) is
extended to define a notion of region-identity over time.
This project has to identified several other major research
areas which we believe merit substantial further
investigation. These include the
importance of
attention within the generalised abductive model, the novel
application of a hybrid point/region based model of visual
space that promises to offer much greater extensibility and
numerical predictive power, to
those provided by the popular region-based formalisms used in
Qualitative Spatial Reasoning; and the development of a general global
workspace model to tackle the challenges of sensor fusion
thrown up by simultaneously integrating data supplied by multiple sensors.
The Cognitive Robotics group
has recently a new 40m2 state-of-the-art robotics laboratory
which is equipped with numerous mobile robot platforms and an
upper-torso humanoid robot. In addition the group has
access to a variety of computer vision equipment, including
three Videre stereoscopic cameras with software, and
pan-and-tilt units as a resource for Cognitive Robotics
research and teaching within the Department of Electrical and
Electronic Engineering.
Central to this project is LUDWIG,
an upper torso, two-armed "humanoid" robot, which was
constructed in the department's engineering
workshops. LUDWIG is a bench mounted with two arms, each of
three degrees-of-freedom. He is equipped with a single
stereoscopic Videre camera mounted on a computer controlled
pan-and-tilt unit from Directed
Perception.
Other robot platforms located in the
laboratory and available for use include: (i) a small fleet of
LinuxBot
robots, each with sonar rangefinders and firewire based
stereoscopic vision which intercommunicate using a
wireless Ethernet LAN; (ii) a Nomad Scout robot (iii) and several
Khepera robots from K-Team and a number of LEGO Mindstorms kits,
available for research and teaching projects.
Closely related
to our project is the work undergone by Yiannis Demiris and
his students researching into human-machine and
machine-machine imitation. The has extended the
selection of available robot platforms with a pair of miniature
walking robots and two PeopleBotTM
mobile robots.
For particularly
computationally intensive tasks, our group has access to
an HP Dual Itanium 2 processor with 4GB RAM, as well as a 32
processor (with planned extensions to 256) "farm".
(l-r
front row) Georgios Sakellariou, Cassandra with Flip, Sunil
Rao, Mercedes Lahnstein; (l-r back row) Mark Witkowski,
Matthew Johnson, David Randell, Murray Shanahan, Ludwig,
Yiannis Demiris, Gavin Simmons, Paschalis Veskos, Adam Rae;
(not available for the photo) Anthony Dearden, Tim Guhl,
Bassam Khadhouri, Seçil Özen, Vidula Vinayagamoorthy
LUDWIG, our humanoid upper-torso robot
A LinuxBot robot with stereoscopic vision
Flip and Flop, for experiments in walking and
imitation
Murray
Shanahan is the principal investigator of this
project, with Mark
Witkowski and David
Randell employed on the project as research
fellows. (Ms.) Seçil Özen, Georgios
Sakellariou and Tim
Guhl, Murray Shanahan's Ph.D. students, are also
engaged in robotics research in the Intelligent Systems and
Networks Group. Past associates employed on previous Cognitive
Robotics projects include Rob Miller
as research fellow with Fabio Berti and Hisashi
Hayashi as research associates. Paulo
Santos,
currently at the University of Leeds, has completed his PhD
with Murray Shanahan and is a past member of the ISN Cognitive
Robotics group.
Dr.
Yiannis Demiris joined the ISN section
from the University of Edinburgh Mobile Robots Group. He is
well known for his work on learning by imitation in robots.
Yiannis Demiris' research covers Robot-Human Interaction;
developmental robotics; biologically-inspired robotics;
computer vision; computational modelling of human brain
perceptual and motor mechanisms with the goal of understanding
the interplay between the mechanisms of action production and
perception, and their pathologies; applications of such brain
models to robotics and robot learning with the goal of
developing robots that can operate and learn in a social,
dynamically-changing environment; development of simulation
and robotic tools that will facilitate such research.
Discussion and correspondence about the work on this
project is very welcome. Please contact Murray Shanahan,
David Randell or
Mark Witkowski
directly.
Papers and Publications by the
Project/ISN Robotics Team
This section presents a selection of papers authored by
members, past and present, of the Imperial College Cognitive
Robotics Group. Other references, and those before 1996, may be found at
the respective author's homepages.
Due to copyright restrictions the text of some recent
papers cannot be placed on the web, please contact one of the
authors directly if you would like a copy, or try the
publisher's web site.
If necessary you can get a copy of the Adobe acrobat reader
for pdf files here.
2004
- Murray Shanahan, An Attempt to Formalise a
Non-Trivial Benchmark Problem in Common Sense Reasoning,
Artificial Intelligence, 153, pages 141-165.
[abstract]
- Murray Shanahan, Perception as Abduction: Turning
Sensor Data into Meaningful Representation, Cognitive
Science, accepted to appear. [abstract]
- Murray Shanahan and David Randell, A Logic-Based
Formulation of Active Visual Perception, in
Principles of Knowledge Representation and Reasoning:
Proceedings of the Ninth International Conference (KR-2004), pages 26-35.
[pdf]
- Jeremy Forth and Murray Shanahan, Indirect and
Conditional Sensing in the Event Calculus, in Proc.
16th European Conference on Artificial Intelligence (ECAI
2004), pages 900-904. [pdf]
- Matthew Johnson and Yiannis Demiris, Abstraction in
Recognition to Solve the Correspondence Problem for Robot
Imitation, in Proc. Towards Autonomic Robotic
Systems (TAROS-04), pages 63-70. [pdf]
- David Randell and Mark Witkowski, Tracking Regions
using Conceptual Neighbourhoods, ECAI-2004,
Workshop on Spatial and Temporal Reasoning, pages
63-71. [pdf]
- Murray Shanahan and Mark Witkowski, Event Calculus
Planning Through Satisfiability, Journal of Logic
and Computation, 14-5, pages 731-745. [abstract]
[jlc]
- Georgios Sakellariou, Murray Shanahan and Benjamin
Kuipers, Skeletonisation as Mobile Robot Navigation,
in Proc. Towards Autonomic Robotic Systems (TAROS-04),
pages 149-155. [pdf]
- Mark Witkowski and Kostas Stathis, A Dialectic
Architecture for Computational Autonomy, in: Nickles,
M., Rovatsos, M. and Weiß, G. (eds.) Agents and
Computational Autonomy - Potential, Risks, and Solutions,
Springer LNCS 2969, pages 261-273. [abstract]
2003
- Paulo Santos and Murray Shanahan, A Logic-base
Algorithm for Image Sequence Interpretation and Anchoring,
Proceedings of the International Joint Conference on
Artificial Intelligence (IJCAI-03), Acapulco, Mexico.
[pdf]
- Yiannis Demiris and Matthew Johnson, Distributed,
Predictive Perception of Actions: a Biologically Inspired
Robotics Architecture for Imitation and Learning, Connection
Science Journal, 15-4, pages 231-243. [pdf]
- Mark Witkowski, Brendan Neville and Jeremy Pitt, Agent
Mediated Retailing in the Connected Local Community, Interacting
with Computers, 15, pages 5-32. [pdf]
- Marcus Santos and Paulo Santos, Sensor Data
Assimilation as Database Transactions, in Papers from
the 2003 AAAI Spring Symposium, Palo Alto, California, 2003,
pages 125-130.
- Paulo Santos, Spatial Reasoning and Abductive
Interpretation of Sensor Data Obtained by a Mobile Robot
in a Dynamic Environment, Ph.D. Thesis, University of
London, Imperial College, Department of Electrical and
Electronic Engineering, September 2003. [pdf]
- Mark Witkowski, Towards a Four Factor Theory of
Anticipatory Learning, in: Butz, M.V., Sigaud, O. and
Gérard, P. (eds.) Anticipatory Behavior in Adaptive
Learning Systems: Foundations, Theories, and Systems,
Springer LNAI 2684, pages 66-85. [pdf]
2002
- Paulo Santos and Murray Shanahan, From Regions to
Transitions, From Transitions to Objects, in AAAI-02
Cognitive Robotics Workshop, in: Baral, C. and
McIlraith, S. (Eds.), Working notes of the AAAI Workshop on
Cognitive Robotics, Edmonton, Canada, 2002 [pdf]
- Paulo Santos and Murray Shanahan, Hypothesising
Object Relations from Image Transitions, in: van Harmelen, F. (Ed.),
15th
European Conference on Artificial Intelligence
(ECAI-02), Lyon, France, July
2002, pages 292-296. [pdf]
- Murray Shanahan, A Logical Account of Perception
Incorporating Feedback and Expectation, in Principles
of Knowledge Representation and Reasoning: Proceedings of
the Eighth International Conference (KR-2002), pages
3-13. [pdf]
- John Demiris and Gillian Hayes, Imitation as a
Dual-route Process Featuring Predictive and Learning
Components: a Biologically-plausible Computational
Model, in: Dautenhahn, K. and Nehaniv, C. (eds.) Imitation in Animals and
Artifacts, MIT Press, Chapter 13. [pdf]
- Mark Witkowski, Anticipatory Learning: The Animat as
Discovery Engine, in: Proc. Adaptive Behavior in
Anticipatory Learning Systems (ABiALS-02), Edinburgh, August
2002 [pdf]
- David Randell and Mark Witkowski, Building Large
Composition Tables via Axiomatic Theories, in
Principles of Knowledge Representation and Reasoning:
Proceedings of the Eighth International Conference
(KR-2002), pages 26-35. [pdf]
2001
- Paulo Santos and Murray Shanahan, From Stereoscopic
Vision to Symbolic Representation, AAAI Fall
Symposium on "Anchoring Symbols to Sensor Data in Single and
Multiple Robot Systems", North Falmouth, MA, pages
37-43. [pdf]
- David Randell, Mark Witkowski and Murray Shanahan,
From Images to Bodies: Modeling and Exploiting Spatial
Occlusion and Motion Parallax, Proc. 17th
IJCAI-01, pages 57-63. This paper was also presented
at the Commonsense
2001 Symposium, New York, May 20-22. [pdf]
- Mark Witkowski, Murray Shanahan, Paulo Santos and David
Randell, Cognitive Robotics: On the Semantic
Knife-edge, Proc.
TIMR 01 - Towards Intelligent Mobile Robots [pdf]
- Mark Witkowski, David Randell and Murray Shanahan,
Deriving Fluents from Sensor Data for Mobile Robots,
AAAI Fall Symposium on "Anchoring Symbols to Sensor Data
in Single and Multiple Robot Systems", North Falmouth,
MA, pages 44-51. [pdf]
- Mark Witkowski, Alexander Artikis and Jeremy Pitt, Experiments
in Building Experiential Trust in a Society of
Objective-Trust Based Agents, in Falcone, R., Singh, M. and Tan, Y-H (eds.) Trust in
Cyber-Society: Integrating Human and Machine Perspective,
Springer LNAI 2246, pages 111-132. [pdf]
2000
- Leliane Nunes de Barros and Paulo E. Santos, The
Nature of Knowledge in an Abductive Event Calculus
Planner, Proceedings of the European Knowledge
Acquisition Workshop (EKAW), Juan les Pins, France. [pdf]
- Murray Shanahan, An Abductive Event
Calculus Planner, The
Journal of Logic Programming, Vol. 44, pages
207-239. [pdf].
- Murray Shanahan and Mark Witkowski, High-Level Robot
Control Through Logic, Proceedings ATAL 2000,
pages 100-113 (to appear in the Springer-Verlag LNAI
series). [pdf]
[appendices]
- Mark Witkowski, The Role of Behavioral Extinction in
Animat Action Selection, Proc 6th Int. Conf. on
Simulation of Adaptive Behaviour (SAB-00), [pdf]
- Mark Witkowski, Alexander Artikis and Jeremy Pitt, Trust
and Cooperation in a Trading Society of Objective-Trust
Based Agents, Proc. Autonomous Agents 2000 Workshop
on Deception, Fraud and Trust in Agent Societies,
pages 127-136 [pdf].
1999
- Hisashi Hayashi, Replanning in
Robotics by Dynamic SLDNF, Working Notes of the
IJCAI 99 Workshop "Scheduling and Planning Meet Real-Time
Monitoring in a Dynamic and Uncertain World", August
1999. [pdf]
- Hisashi Hayashi, Abductive Constraint Logic
Programming with Constructive Negation, in the
Working Notes of the Third
Workshop on Non-Monotonic Reasoning, Action, and
Change, International Joint Conference on
Artificial Intelligence, August 1999. [pdf]
- Murray Shanahan, A Logical Account
of the Common Sense Informatic Situation for a Mobile
Robot, Electronic Transactions on
Artificial Intelligence. [pdf]
- Murray Shanahan, The Ramification
Problem in the Event Calculus, Proceedings IJCAI
99, pages 40-46. [pdf]
- Murray Shanahan, What Sort of Computation Mediates
Best between Perception and Action? Logical
Foundations for Cognitive Agents: Contributions in Honor of
Ray Reiter, ed. H.Levesque and F.Pirri, Springer-Verlag,
pages 352-369. [pdf]
- Mark Witkowski, Integrating
Unsupervised Learning, Motivation and Action Selection in an
A-Life Agent, Proceedings
5th European Conf. On Artificial Life
(ECAL99), September 1999. [pdf]
- Mark Witkowski, Applying
Unsupervised Learning and Action Selection to Robot
Teleoperation, Towards
Intelligent Mobile Robots (TIMR-99), Bristol, March
1999. [pdf]
1998
- Hisashi Hayashi, Knowledge
Assimilation and Proof Restoration through the Addition of
Goals, Proceedings
8th International Conference on Artificial Intelligence:
Methodology, Systems, and Applications (AIMSA98),
Springer-Verlag LNAI 1480, pages 291302.
- A G Cohn, N M Gotts, Z Cui, D A Randell, B Bennet and J
M Gooday, Exploiting Temporal Continuity in Qualitative
Spatial Calculi, in: Egenhofer, M.J. and Golledge, R.
(eds.),
Spatial Information Systems, Oxford, pages 5-24. [pdf]
- Murray Shanahan, Reinventing
Shakey, Working Notes of the 1998 AAAI Fall
Symposium on Cognitive Robotics, pages 125135. [pdf]
- Mark Witkowski, Dynamic Expectancy:
An Approach to Behaviour Shaping Using a New Method of
Reinforcement Learning, 6th Int.
Symp. on Intelligent Robotic Systems (SIRS98), July,
1998, pages 73-81. [pdf]
1997
- Hisashi Hayashi, Language HSimple(R):
An Action Language for Representing Concurrent Actions and
Continuous Changes, in the Working Notes of the
Second
Workshop on Practical Reasoning and Rationality, [pdf]
- Murray Shanahan, Solving the Frame Problem: A
Mathematical Investigation of the Common Sense Law of
Inertia, MIT
Press. [contents]
- Murray Shanahan, Event Calculus Planning
Revisited, in Proceedings of 1997 European Conference
on Plannning (ECP 97). An early draft of this paper
appears in the Working Notes of the AAAI 97 Workshop on
Robots, Softbots, Immobots: Theories of Action, Planning and
Control, 1997. [pdf]
- Murray Shanahan, Noise,
Non-Determinism and Spatial Uncertainty,
Proceedings AAAI 97, pages 153158. [pdf]
1996
- Rob Miller and Murray Shanahan, Reasoning about
Discontinuities in the Event Calculus, in proceedings of
the 5th International Conference on Principles of Knowledge
Representation and Reasoning (KR 96), pages 63-74. [pdf]
- Murray Shanahan, Noise and the Common Sense
Informatic Situation for a Mobile Robot, in the
proceedings of AAAI'96, pages 1098-1103. [pdf]
- Murray Shanahan, Robotics and the Common Sense
Informatic Situation, in the proceedings of ECAI'96,
Budapest, Hungary, pages 684-688. [pdf]
- The
Intelligent Systems and Networks Section at Imperial
College
- Artificial
Intelligence Laboratory, MIT
- Artificial
Intelligence Laboratory, University of Texas at Austin
- Autonomous
Agents Laboratory for Cognitive Robotics (ALCOR) Group,
University of Rome, "La Sapienza"
- Cognitive
Robotics Group, University of Toronto, Canada
- Computer
Science Department, University of Rochester
- Computer
Science and Engineering, Arizona State University
- Declarative
Languages and Artificial Intelligence (DTAI), University
of Leuven
- Department of
Computer Science, University of Exeter
- Department of
Computing, Imperial College, London
- Faculty of
Engineering at the University of the West of England (UWE)
- Fraunhofer
Institute for Intelligent Systems (AIS)
- Institute
of Computer Science, University of Bonn
- Intelligent
Robotics Group, University of Wales, Aberystwyth
- Knowledge-Based
Systems Group, RWTH, Aachen
- Knowledge
Representation and Reasoning, University of Leeds
- Laboratory
for Cognitive Autonomous Systems, University of Linköping
- Mobile
Robot Programming Laboratory, Carnegie Mellon
- Mobile
Robotics Laboratory at AAAS, University of Örebro
- Robotics
Research at the University of Essex
- School of
Informatics, University of Edinburgh
- Stanford AI
Laboratory
- Stanford
Computer Science (John McCarthy)
- Wasada
University Humanoid Project
This page maintained by Mark Witkowski. Last
Change: 21/10/04 (by MW).
|
