This paper explains in detail a biologically inspired encoding scheme for the artificial evolution of neural network robot controllers. Under the scheme, an individual cell divides and moves, in response to protein interactions with an artificial genome, to form a multi-cellular `organism'. After differentiation dendrites grow out of each cell, guided by chemically sensitive growth cones, to form connections between the cells. The resultant network is then interpreted as a recurrent neural network robot controller. Results are given of preliminary experiments to evolve robot controllers for both corridor following and object avoidance tasks. Keywords: Artificial Evolution, Morphogenesis, Evolutionary Robotics. 1 Introduction Many people have noted the difficulties involved in designing control architectures for robots by hand ([10], [1]) and as robots and the behaviours we demand of them become more complicated these difficulties can only increase. Evolutionary robotics is an attempt...

This paper provides a high-level review of current and recent work in the use of genetic algorithm based techniques to develop sensorimotor control systems for autonomous agents. It focuses on network-based controllers and genetic encoding issues associated with them. The paper closes with a discussion of the possibility of using arti cial evolutionary techniques to help tackle more specifically scientific questions about natural sensorimotor systems.

I draw a distinction between orthodox cognitive science and biological cognitive science. The former tends to ignore biological considerations whilst the latter holds that life and mind share a common set of organizational principles. The suggestion here is that artificial life (A-Life) is (potentially) the intellectual engine of the latter. The goal then becomes to map out the conceptual profile of that A-Life-driven cognitive science. Paying special attention to the relationship between neurobiological/biochemical phenomena and cognition, I argue that the commitment to functionalism in orthodox cognitive science provides compelling evidence that that approach is wedded to a recognizably Cartesian account of the relationship between life and mind. By contrast, the fundamental commitments of a biological cognitive science tell in favour of a radically different, generically Aristotelian framework. I show how the concept of self-organization --- arguably the central theoretical idea in...

The generation of complex entities with advantageous behaviours beyond our manual design capability requires long-term incremental evolution with continuing emergence. In this paper, we argue that artificial selection models, such as traditional genetic algorithms, are fundamentally inadequate for this goal. Existing natural selection systems are evaluated, revealing both significant achievements and pitfalls. Thus, some requirements for the perpetuation of evolutionary emergence are established. An (artificial) environment containing simple virtual autonomous organisms with neural controllers has been created to satisfy these requirements and to aid in the development of an accompanying theory of evolutionary emergence. Resulting behaviours are reported alongside their neural correlates. In a particular example, the collective behaviour of one species provides a selective force which is overcome by another species, demonstrating the incremental evolutionary emergence of advantageous behaviours via naturally-arising coevolution. While the results fall short of the ultimate goal, experience with the system has provided some useful lessons for the perpetuation of emergence towards increasingly complex advantageous behaviours.

Many of the techniques commonly used to evolve neural networks for robots face various problems if the size and morphology of these networks are allowed to vary under evolutionary control. This paper identifies some of these problems for commonly used encoding schemes, neural networks, genetic operators and evaluation techniques and proposes some new techniques that may be used to alleviate them. Results from experiments are reported in which these new techniques were combined for the open-ended evolution of a many-faceted robot behaviour.

The evolution of complexity is investigated in the context of an `iterated prisoner's dilemma' (IPD) co-evolutionary/game-theoretic ecology, populated by strategies determined by variable length genotypes. New evidence is found to support the dual hypotheses that both uncertainty, and interaction (by way of population stability), foster the evolution of progressively more complex entities. It is also argued that during periods of major evolutionary upheaval, complex entities suffer disproportionately and become less abundant in the population. The research is presented as an elaboration of the general principle that there is complexity in an organism by virtue of complexity in the environment, and has implications both for deepening understanding of the nature of biological evolution and for guiding the progress of artificial evolution. 1 Introduction As Stephen Jay Gould [6] has consistently pointed out, the age of bacteria is not about to end anytime soon. Yet it can hardly be deni...

This paper introduces an integration of reinforcement learning and behavior-based control designed to produce real-time learning in situated agents. The model layers a distributed and asynchronous reinforcement learning algorithm over a learned topological map and standard behavioral substrate to create a reinforcement learning complex. The topological map creates a small and task-relevant state space that aims to make learning feasible, while the distributed and asynchronous aspects of the architecture make it compatible with behavior-based design principles.

Identification of the fundamental properties necessary for the generation of adaptive behaviour is one of the primary goals for Artificial Life. In this paper, we address the related question of whether we can identify general useful properties of a given solution class. Such an approach provides a potentially scalable framework that may enable us to identify general properties of more complex adaptive systems. We develop a methodology based on analysis of successfully evolved solutions to an evolutionary robotics shape discrimination problem, allowing us to identify properties of solution classes that are potentially useful over a wider class of problems than the original task. We propose that the evolvability of the solution class is due to the fundamental property of temporal adaptivity.

This paper argues that evolutionary robotics (ER) techniques can act as useful and potentially wide ranging tools in the scientific investigation of adaptive behaviour. After discussing the kinds of investigations ER can play a central role in, a concrete example is presented. We conclude that these kinds of studies are not only scientifically useful, but are necessary for the field to develop as an engineering methodology for autonomous robotics. 1 Introduction Evolutionary robotics (ER) involves evaluating, over a number of generations, whole populations of autonomous robot control systems specified by artificial genotypes. These are interbred using a Darwinian scheme in which the fittest individuals are most likely to produce offspring. Fitness is measured in terms of how good a robot's behaviour is according to some task-based evaluation criterion. This particular flavour of new-wave autonomous robotics was originally proposed as an automatic alternative to hand design of control...

Evolutionary search algorithms (ESAs from now on) are iterative problem solvers developed with inspiration from neo-Darwinian survival of the fittest genes. This thesis looks at the core issues surrounding ESAs and is a step towards building a rational methodology for their effective use. Currently there is no such method of best practice rather the whole process of designing and using ESAs is seen as more of a black art than a tried and tested engineering tool. Consequently, many non-practitioners are sceptical of the worth of ESAs as a useful tool at all. Therefore the first task of the thesis is to lay out the reasons, from computational theory, why ESAs can be a potentially powerful tool. In this context the theory of NP-completeness is introduced to ground the discussions throughout the thesis. Then a generic framework for describing ESAs is developed to form another cornerstone of these later discussions. As well as arguing for their potential power, the main argument of the the...