i) it is consistent with biological data and may constitute a model of biological spatial memory,
ii) it is completely functional in the sense that pre- and post-processing elements are sufficiently detailed to enable a real world implementation supporting useful spatial behaviour in robots.
Progress Report:
The project is still at an early stage. Lessons are currently being drawn from past investigations in the field. Those involved first grid based planning models (Bugmann, Taylor, Denham, 1994a, 1994b; Bugmann, 1996). Later a model based on learning of action-perception-action triplets was developed and presented at ASI-AA-95, Ascona. The role of sequence learning was also explored in (Denham and McCabe, 1995, 1996a, 1996b). The latest model is a view-based model in which view-nodes are associated with objects, and actions and are linked according to the history of the exploration. Planning proceeds by activity spread in view-nodes. A book chapter describes this model and recommendations for the design of artificial spatial memory models (Bugmann, 1997). Essentially, it is suggested that view-based models are too brittle to be of practical use.
Past work involved also the supervision of students in using vision for robotics (e.g. Onillon et al., 1995) and a collaboration with Prof. Jack Boitano from Fairfield University, USA (Boitano et al., 1996) in investigating the function of the Hippocampus. Other publications related to vision and the function of biological neurons can be found in the home page of G. Bugmann
A special session on
Biologically Inspired Control
was organised at the IEE Conference Control'96 in Exeter.
With funding of the Henri Moore Fundation, the group has acquired a motorised
wheelchair an provided it with a computerised control system for the exhibition
of an
autonomous wheelchair
in an Art Gallery. A neural network based on normalised RBF nets was
used for encoding the trajectory of the wheelchair (Bugmann et al., 1998).
This wheelchair is now available for model testing.
A Rug Warrior Robot is used, provided with wireless RS232 communication,
video camera, and TV transmitters. Image processing and neurocomputing is
performed on a PC using the neural network simulation package CORTEX-PRO
with extended I/O functions.
The PhD study by Kheng Lee Koay
is close to completion. He has built a complete system that uses vision for
self-localization and obstacle detection. Planning is performed using a
resistive grid. A modified version of the Smith Predictor has been
designed to manage the time delays and intermitency in the visual feedback
loop.
A collaboration has been initiated with
Philippe Gaussier
at the ENSEA in Paris who spend two weeks in our group in 1996. Current
work reviews behavioural data on planning processes in the human, and attempts
to produce a model of the anatomical basis for planning.
Work on the role of the cerebellum in planning when there is limited time to react, such as in a tennis game, hase been presented at the IEE workshop " Self-learning robots III Brainstyle robotics: The cerebellum beyond function approximation ". It is now published as a chaper in LNAI 2036 (Bugmann, 2001).
A current EPSRC project explores the acquisition of route knowledge
via verbal instructions . This requires the design of high-level navigation
primitives. More information is available on the
IBL projec
t webpage.
References:
"Role of the cerebellum in time-critical goal-oriented behaviour:
Anatomical basis and control principle." (Preprint: ps.zip (250.872)
PDF (111.833))
Bugmann G. (2001)
in Wermter S., Austin J. and Willshaw D. (eds) “Emergent Neural Computational
Architectures Based on Neuroscience”, Lecture Notes in Computer Science (LNAI
2036), Springer, March 2001, pp. 295-273 (available on-line from Spring-Verlag
).(ISBN: 3-540-42363-X)
"Stable Encoding of Robot Trajectories using Normalised Radial Basis Functions:
Application to an Autonomous Wheelchair" (83,209)
,
Bugmann G., Koay K.L., Barlow N., Phillips M. and Rodney D. (1998)
Proc. 29th Intl. Symp. Robotics, 27-30 April, Birmingham, UK.
"A Connectionist Approach to Spatial Memory and Planning" (167,080)
Bugmann, G. (1997),
as Chap. 5 in Landau L.J. and Taylor J.G. (eds) "Basic Concepts in Neural
Networks: A survey",
In the Series: Perspectives in Neural Networks, Springer, London, pp.
109-146.
"Effect of Medial Septal Lesions:
Implications for Models of Hippocampal Function"
Boitano J.R., Bugmann G., Bapi R.S., McCabe S.L., Dokla C.P.J. and Denham
M.J. (1997)
in: Bower J. (ed), Computational Neuroscience, Plenum Press, NY, pp. 577-583.
(Proc. of CNS'96
, Boston)
"Value Maps for planning and
learning implemented with cellular automata "
Bugmann G. (1996)
in Parmee I.C. (ed) "Proc. of the 2nd International Conf. on Adaptive
Computing in Engineering Design and Control (ACEDC'96)", Plymouth, 26-28
March 1996, ISBN 0 905227 61 1, p. 307-309.
edc96_ps.zip (30631)
"Biological temporal sequence
processing and its application in robot control"
Denham M.J. and McCabe S.L. (1996a)
Proc. CONTROL'96, Exeter, UK, September 1996.
"Biological basis for a neural
model of learning and recall of goal-directed sensory-motor behaviours"
Denham M.J. and McCabe S.L. (1996b)
Proc World Congress on Neural Networks (WCNN'96), San Diego, USA, September
1996.
"Artificial vision for micromouse"
Onillon V., Bugmann G., Simpson A. and Nurse P. (1995) Research Report
NRG-95-05, School of Computing, University of Plymouth, Plymouth PL4 8AA,
UK.
nrg9505.zip (ps file)(80620)
"Robot control using temporal
sequence learning"
Denham M.J. and McCabe S.L. (1995)
Proc. World Congress on Neural Networks (WCNN'95), Washington D.C., USA,
"Route finding by neural nets"
Bugmann, G., Taylor, J.G. and Denham M. (1994)
in Taylor J.G (ed) "Neural Networks", Alfred Waller Ltd, Henley-on-Thames,
p. 217-230.
"Sensory and Memory-Based Path-Planning
in the Egocentric Reference Frame of an Autonomous Mobile Robot"
Bugmann G, Taylor J. G., and Denham M. J. (1994)
Research Report NRG-94-01, School of Computing, University of Plymouth,
Plymouth PL4 8AA, UK.
nrg9401.zip (ps file)(587308)