Evolutionary Robotics - Tech. Reports

[Inman Harvey]

Evolutionary Robotics at Sussex -- Technical Reports
===============================

The following six technical reports describe our recent work in using genetic
algorithms to develop neural-network controllers for a simulated simple
visually-guided robot.

Currently only hard-copies are available. To request copies, mail one of:
inmanh at cogs.susx.ac.uk or davec at cogs.susx.ac.uk or philh at cogs.susx.ac.uk
giving a surface mail address and the CSRP numbers of the reports you want.

or write to us at:
School of Cognitive and Computing Sciences
University of Sussex
Brighton BN1 9QH
England, UK.

------------ABSTRACTS--------------------

Genetic convergence in a species of evolved robot control architectures
I. Harvey, P. Husbands, D. Cliff
Cognitive Science Research Paper CSRP267
February 1993
We analyse how the project of evolving 'neural' network controller for
autonomous visually guided robots is significantly different from the usual
function optimisation problems standard genetic algorithms are asked to
tackle.  The need to have open ended increase in complexity of the
controllers, to allow for an indefinite number of new tasks to be
incrementally added to the robot's capabilities in the long term, means that
genotypes of arbitrary length need to be allowed. This results in
populations being genetically converged as new tasks are added, and needs a
change to usual genetic algorithm practices. Results of successful runs are
shown, and the population is analysed in terms of genetic convergence and
movement in time across sequence space.


Analysing recurrent dynamical networks evolved for robot control
P. Husbands, I. Harvey, D. Cliff
Cognitive Science Research Paper CSRP265
January 1993
This paper shows how a mixture of qualitative and quantitative analysis can
be used to understand a particular brand of arbitrarily recurrent continuous
dynamical neural network used to generate robust behaviours in autonomous
mobile robots.  These networks have been evolved in an open-ended way using an
extended form of genetic algorithm.  After briefly covering the background to
our research, properties of special frequently occurring subnetworks are
analysed mathematically. Networks evolved to control simple robots with low
resolution sensing are then analysed, using a combination of knowledge of
these mathematical properties and careful interpretation of time plots of
sensor, neuron and motor activities.

Analysis of evolved sensory-motor controllers
D. Cliff, P. Husbands, I. Harvey
Cognitive Science Research Paper CSRP264
December 1992
We present results from the concurrent evolution of visual sensing
morphologies and sensory-motor controller-networks for visually guided robots.
In this paper we analyse two (of many) networks which result from using
incremental evolution with variable-length genotypes. The two networks come
from separate populations, evolved using a common fitness function. The
observable behaviours of the two robots are very similar, and close to the
optimal behaviour. However, the underlying sensing morphologies and
sensory-motor controllers are strikingly different. This is a case of
convergent evolution at the behavioural level, coupled with divergent
evolution at the morphological level. The action of the evolved networks is
described. We discuss the process of analysing evolved artificial networks, a
process which bears many similarities to analysing biological nervous systems
in the field of neuroethology.

Incremental evolution of neural network architectures for adaptive behaviour
D. Cliff, I. Harvey, P. Husbands
Cognitive Science Research Paper CSRP256
December 1992
This paper describes aspects of our ongoing work in evolving recurrent
dynamical artificial neural networks which act as sensory-motor controllers,
generating adaptive behaviour in artificial agents. We start with a discussion
of the rationale for our approach. Our approach involves the use of recurrent
networks of artificial neurons with rich dynamics, resilience to noise (both
internal and external); and separate excitation and inhibition channels. The
networks allow artificial agents (simulated or robotic) to exhibit adaptive
behaviour. The complexity of designing networks built from such units
leads us to use our own extended form of genetic algorithm, which allows for
incremental automatic evolution of controller-networks. Finally, we review
some of our recent results, applying our methods to work with simple
visually-guided robots. The genetic algorithm generates useful network
architectures from an initial set of randomly-connected networks. During
evolution, uniform noise was added to the activation of each neuron. After
evolution, we studied two evolved networks, to see how their performance
varied when the noise range was altered. Significantly, we discovered that
when the noise was eliminated, the performance of the networks degraded: the
networks use noise to operate efficiently.

Evolving visually guided robots
D. Cliff, P. Husbands, I. Harvey
Cognitive Science Research Paper CSRP220
July 1992
We have developed a methodology grounded in two beliefs: that autonomous
agents need visual processing capabilities, and that the approach of
hand-designing control architectures for autonomous agents is likely to be
superseded by methods involving the artificial evolution of comparable
architectures. In this paper we present results which demonstrate that
neural-network control architectures can be evolved for an accurate simulation
model of a visually guided robot. The simulation system involves detailed
models of the physics of a real robot built at Sussex; and the simulated
vision involves ray-tracing computer graphics, using  models of optical
systems which could readily be constructed from discrete components. The
control-network architecture is entirely under genetic control, as are
parameters governing the optical system. Significantly, we demonstrate that
robust visually-guided control systems evolve from evaluation functions which
do not explicitly involve monitoring visual input. The latter part of the
paper discusses work now under development, which allows us to engage in
long-term fundamental experiments aimed at thoroughly exploring the
possibilities of concurrently evolving control networks and visual sensors for
navigational tasks. This involves the construction of specialised
visual-robotic equipment which eliminates the need for simulated sensing.

Issues in evolutionary robotics
I. Harvey, P. Husbands, D. Cliff
Cognitive Science Research Paper CSRP219
July 1992
In this paper we propose and justify a methodology for the development of
the control systems, or `cognitive architectures', of autonomous mobile
robots. We argue that the design by hand of such control systems becomes
prohibitively difficult as complexity increases. We discuss an alternative
approach, involving artificial evolution, where the basic building blocks for
cognitive architectures are adaptive noise-tolerant dynamical neural networks,
rather than programs. These networks may be recurrent, and should operate in
real time. Evolution should be incremental, using an extended and modified
version of genetic algorithms. We finally propose that, sooner rather than
later, visual processing will be required in order for robots to engage in
non-trivial navigation behaviours. Time constraints suggest that initial
architecture evaluations should be largely done in simulation. The pitfalls of
simulations compared with reality are discussed, together with the importance
of incorporating noise. To support our claims and proposals, we present
results from some preliminary experiments where robots which roam office-like
environments are evolved.