Search

This paper proposes a paradigm for specification of virtual human agents' level of autonomy. The idea we present in this paper aims at optimi sing the required complexity of agents in order to perform realistic simulations. The paradigm is based on the distribution of the required virtual human agent "intelligence" to other simulation entities like groups of agents, and objects. [ABSTRACT FROM AUTHOR]

A framework for understanding and exploiting embodiment is presented which is not dependent on any specific ontological context. This framework is founded on a new definition of embodiment, based on the relational dynamics that exist between biological organisms and their environments, and inspired by the structural dynamics of the bacterium Escherichia coli. Full recognition is given to the role played by physically instantiated bodies, but in such a way that this can be meaningfully abstracted within the constraints implied by the term ‘embodiment', and applied in a variety of operational contexts. This is illustrated by ongoing experimental work in which the relational dynamics that exist between E. coli and its environment are applied in a variety of software environments, using Cellular Automata (CA) with artificial' sensory' and' effector' surfaces, producing qualitatively similar' chemotactic' behaviours in a variety of operational domains. [ABSTRACT FROM AUTHOR]

Nap-of-the-Earth (NOE) flight and the development of advanced Unmanned Air Vehicles (UAV) require further integration of sensors and onboard processing. Flying insects feature a panoramic eye which relies on Optical Flow (OF) for guidance and obstacle avoidance. We describe the development of a miniature thrust-vectoring UAV featuring an analog electronic eye based on biological motion detection. The system is tested within the laboratory on a custom-built whirling-arm rig. Preliminary tests have demonstrated satisfactory maneuvrability under remote-control. The project combines the fields of neurobiology, robotics, and aerospace. A VLSI implementation of the vision system is planned. [ABSTRACT FROM AUTHOR]

The design of behavior generating control structures for real robots acting autonomously in a real and changing environment is a complex task. This is in particular true with respect to the debugging process, the documentation of the encountered behavior, its quantitative analysis and the final evaluation. To successfully implement such a behavior, it is vital to couple the synthesis on a simulator and the experiment on a real robot with a thorough analysis. The available simulator tools in general only allow behavioral snapshots and do not provide the option of online interference. In order to cure these shortcomings, a visualization tool for aposteriori graphical analysis of recorded data sets which gives access to all relevant internal states and parameters of the system is presented. The mini-robot Khepera has been chosen as experimentatory platform. [ABSTRACT FROM AUTHOR]

In this work we present a methodology for the progressive construction of compound behavior controllers for real autonomous robots. Some of these behaviors require temporal processing which is achieved through the inclusion of temporal delays in the synapses of the artificial neural networks used for their implementation. Starting from a set of simple behaviors implemented by means of evolved monolithic controllers, the evolution strategy employed obtains behaviors in higher levels either choosing the necessary low level behaviors from the previously selected set or through the coevolution of part of the low level behaviors and the higher level one. Emphasis is placed on making the behaviors robust and capable of performing in a real robot. [ABSTRACT FROM AUTHOR]

We present a neural model for the control of an animat. The model is based on two structures. The first one enables visual navigation using landmarks. It may be used in unknown and changing environments. The second structure enables building a proximity map of the environment. Using this map, an animat may successfully reach different goals linked to different motivations and solve various types of action selection problems. [ABSTRACT FROM AUTHOR]

In this paper, we present a neural architecture for a mobile robot in order to learn how to imitate a sequence of actions. We show that the use of a representation of the information in a continuous and dynamic way is necessary and the use of the neural fields can be a good solution to control the dynamic of several degrees of freedom with a single internal representation. [ABSTRACT FROM AUTHOR]

The paper gives probably the first ever systematic discussion on how wave processes in reaction-diffusion and excitable homogeneous media can be efficiently used to solve a wide range of problems in robot navigation. Three possible applications of chemical controllers are considered: (i) object following, (ii) optimal path finding, and (iii) universal control. The various implementations of controllers discussed here include: Belousov-Zhabotinsky chemical processors, families of excitable lattices, and self-localised excitations. We present some results from a simulation of robot control using excitable lattices, and find the results encouraging for our planned construction of a chemically controlled robot. [ABSTRACT FROM AUTHOR]

This paper proposes novel adaptation method for a behavior-based locomotion robot. Utilization of hierarchical behavior controller makes the controller designing process easier and shorter, because the designer can deal with behavior controllers for simple behaviors independently, and coordinate those behavior controllers in order for a robot to perform the objective behavior after finishing the design of the fundamental behavior controllers. Some problems are still remaining. One is how to adjust the behavior coordinator when the objective behavior or robot parameters are slightly changed. We propose the novel method to adjust the behavior coordinator against some changes. This method measures the effects of the fundamental behavior controllers to the total behavior, and changes the activation values for them in less trials. This proposed method is applied to the real brachiation robot control. This brachiation robot has redundant mechanism to locomote from branch to branch with 14 actuators like a long-armed ape. [ABSTRACT FROM AUTHOR]

With a few exceptions, today's mobile robots, however complex, are not truly autonomous. At some time, they all require humans to supply them with energy and/or information; most also require other forms of assistance. In complete contrast, even the simplest animals are totally self-sufficient. We describe a current project1 which aims to construct autonomous robots with animal-like self-sufficiency both in terms of energy and information. The robots will live free on agricultural land, hunting and catching slugs, and fermenting the corpses to produce biogas, which will fuel the generator providing the robots with power. [ABSTRACT FROM AUTHOR]

In this paper, the processes of exploration and of incremental learning in the robot navigation task are studied using the dynamical systems approach. A neural network model which performs the forward modeling, planning, consolidation learning and novelty rewarding is used for the robot experiments. Our experiments showed that the robot repeated a few variation of travel patterns in the beginning of the exploration, and later the robot explored more diversely in the workspace by combining and mutating the previously experienced patterns. Our analysis indicates that internal confusion due to immature learning plays the role of a catalyst in generating diverse action sequences. It is found that these diverse exploratory travels enable the robot to acquire the rational modeling of the environment in the end. [ABSTRACT FROM AUTHOR]

This paper describes an evolutionary way to design behaviors of a mobile robot for recognizing environments. We have proposed an action-based approach (called AEM) for a mobile robot to recognize environments. In AEM, a behavior-based mobile robot acts in each environments and action sequences are obtained. The action sequences are transformed into vectors characterizing the environments, and the robot identifies the environments with the vectors. The design of suitable behaviors for AEM is very difficult for human because the search space is huge and intuitive understanding is hard. Thus we develop the evolutionary design of such behaviors using genetic algorithm. [ABSTRACT FROM AUTHOR]

This paper describes the generation and selection of visual feature detectors. The feature detectors are randomly generated, and are built out of components, some having functionality inspired on observations of animal visual pathways. The input for the feature detectors consists of non-synthetic images, while the selectionist pressure comes from the amount of information the feature detectors generate. The experimental setup is described and some results are given. [ABSTRACT FROM AUTHOR]

In this paper we describe our attempt to create a nature-like simulation model of artificial creatures. The model includes physical simulation of creatures, their interaction with the environment, their neural network control, and both directed and open-ended evolution. We describe a complex, three-dimensional simulation system, where various fitness criteria can be selected for evolving species, and a spontaneous evolution can be run. The work is still being developed, and we hope to make it a realistic model capable of producing real-life phenomena through an open-ended evolution in a life-like world of stick creatures. [ABSTRACT FROM AUTHOR]

The relationship between evolution (genetic & developmental processes of an evolutionary system) and modularity (its support for modular structures) is explored. Modules are defined as structures with common origin; either evolutional or developmental. In the former case, processes supporting modularity operate on the phylogenetic level, in the latter, on the ontogenetic level. Three such processes are identified (duplication, divergence, covergence). The existence of these processes determine the system's support for modularity. Modules are analysed in the particular context of artificial neural networks (ANNs), where they appear as subnetworks. Gruau's cellular developmental encoding is used as an example, and an extension is proposed which better supports modularity. [ABSTRACT FROM AUTHOR]

Using evolutionary simulations, we develop autonomous agents controlled by artificial neural networks (ANNs). In simple life-like tasks of foraging and navigation, high performance levels are attained by agents equipped with fully-recurrent ANN controllers. Examining several experimental settings, differing in the sensory input available to the agents, we find a common structure of a "command neuron" switching the dynamics of the network between radically different behavioural modes. In some of the models the command neuron reflects a map of the environment, acting as a "place cell". In others it is based on a spontaneously evolving short-term memory mechanism. The resemblance to known findings from neurobiology places Evolved ANNs as an excellent candidate model for the study of structure and function relation in complex nervous systems. [ABSTRACT FROM AUTHOR]

The object of this work is to examine the early evolution of the nervous system in relation to adaptive behavior. The main questions are: how the nervous system came into being, how a system can be organized during evolution that is able to ensure the adaptive behavior of a being, what are the basic rules of construction that are sufficient to create a workable nervous system without specifying the details of the construction. The biological bases of the model are the phyla Cnidaria and Porifera because they stand at the beginning of the genesis of nervous organization. We found that if during the evolutionary process a kind of cell comes into being that is able to conduct electrical stimuli - even in a rudimentary way - than this kind of cell improves the behavioral performance by itself without containing any specific information of how to organize the construction of this system. [ABSTRACT FROM AUTHOR]

To be useful in psychology "artificial organisms" have to perform tasks comparable to those performed by animals. One way to achieve this is to replicate actual animal experiments. Here we reproduce an experiment showing "detour behavior" in chicks - a behavior usually explained in terms of "cognitive maps" or other forms of internal representation. We artificially evolve software-simulated robots with a "generic" ability to detour. Sensor-motor physics are carefully calibrated with data from a physical robot. Robot architecture is constrained to exclude internal representation. The evolutionary process rewards exploratory skills as well as detour behavior. Robot performance matches the results achieved in the original experiment. This proves that internal representations are not a necessary condition for primitive detour behavior and suggests that "detouring" evolves naturally from simpler behaviors. Future research will show whether it is possible to evolve more complex detour abilities using a similar bottom-up strategy. [ABSTRACT FROM AUTHOR]

In this paper we introduce the notion of historical evidence — the ability to replicate biologically realistic evolutionary scenarios — for hypothesised mechanisms for control of sensorimotor behaviour. We apply the idea to the phonotaxis mechanism proposed by Webb and her collaborators to account for the abilities of Gryllus bimaculatus. To do this, we tested whether the proposed control mechanism, when implemented in a robot model of the animal, could account for evolutionary adaptations observed in the natural system. We describe and discuss the experiment and its results, but start by explaining the methodology used, which is an extension of Webb's existing methodology for obtaining behavioural evidence for a hypothesised mechanism. We conclude there is historical evidence for the neural control mechanism investigated. [ABSTRACT FROM AUTHOR]

This paper briefly reviews synthetic approaches to neurobiology and presents results of two experiments on the use of evolutionary algorithms for the design of neural controllers for locomotion. The first experiment consists in using the evolutionary algorithm for instantiating low level parameters of a connectionist simulation of the lamprey's locomotor circuitry. The second experiment develops potential neural circuits for the swimming and trotting of the salamander; an animal whose locomotor circuitry has currently not been decoded. In both cases, biologically plausible control circuits are developed which produce a neural activity with many similarities to that measured in the real animals. [ABSTRACT FROM AUTHOR]

This paper is concerned with artificial evolution of neurocontrollers with adaptive synapses for autonomous mobile robots. The method consists of encoding on the genotype a set of local modification rules that synapses obey while the robot freely moves in the environment [2]. The synaptic weights are not encoded on the genotype. In the experiments presented here, a "behavior-based fitness" function gives reproductive advantage to robots that can solve a sequential task. The results show that evolutionary adaptive controllers solve the task much faster and better than evolutionary standard (non-adaptive) controllers, that the method scales up well to large architectures whereas standard controllers do not, and that evolved adaptive controllers are not trivial and cannot be reduced to a fixed-weight network. [ABSTRACT FROM AUTHOR]

One important implication of embodiment is that, by acting, agents partially determine the sensory patterns they receive from the environment. The motor actions performed by an agent, by modifying the agent's position with respect to the external environment and/or the external environment itself, partially determine the type of sensory patterns received from the environment In this paper we investigate how agents can take advantage of this ability. In particular, we discuss how agents coordinate sensory and motor processes in order to (1) select sensory patterns which are not affected by the aliasing problem and avoid those which are; (2) select sensory patterns such that groups of patterns which require different responses do not strongly overlap; (3) exploit emergent behaviors that result from the interaction between the agent and the environment. [ABSTRACT FROM AUTHOR]

We study the evolution of computation performed by non-uniform cellular automata in which global information processing appears at two different levels of self-organisation. In our model, the first level of self-organisation is characterised by interactions among cellular macrostructures or computational demes which compete for room in a finite grid of cells. This level is related to the formation, evolution and extinction of macrostructures, and it is designed in a completely local manner. The second level of self-organisation refers to the interactions among the cells within the demes. The model, derived from the cellular programming approach, allows global computation to occur as a result of many local interactions among computational demes of interacting cells. The study reveals some of the mechanisms by which co-evolving demes of non-uniform cellular automata perform non-trivial computation, such as the synchronisation tasks. [ABSTRACT FROM AUTHOR]

This report presents an asynchronous, distributed genetic programming (GP) system using a master/slave architecture. Using symbolic regression for fourier functions as the problem domain, the system was found to demonstrate cooperative coevolutionary dynamics when multiple client populations evolve solutions to similar, but different problems: specifically, closely coupled populations were found to promote continuous search, which in some cases leads to the discovery of better solutions. [ABSTRACT FROM AUTHOR]

Grammatical Evolution (GE) is a grammar-based GA which generates computer programs. GE has the distinction that its input is a BNF, which permits it to generate programs in any language, of arbitrary complexity. Part of the power of GE is that it is closer to natural DNA than other Evolutionary Algorithms, and thus can benefit from natural phenomena such as a separation of search and solution spaces through a genotype to phenotype mapping, and a genetic code degeneracy which can give rise to silent mutations that have no effect on the phenotype. It has previously been shown how runs of GE are competitive with GP, and in this paper we analyse the feature of genetic code degeneracy, and its implications for genotypic diversity. Results show that genetic diversity is improved as a result of degeneracy in the genetic code for the problem domains addressed here. [ABSTRACT FROM AUTHOR]

Since the early beginnings of Evolutionary Computation, Finite State Machines (FSMs) have been applied to model organisms. We present a new approach to evolve such artificial organisms. The FSMs are subject to a difficult navigation and searching task in heterogeneous environments. We give a definition of FSM-species and investigate their formation. The results show that species are formed as the organisms agree on a common ‘genetic broadcast language' and take advantage of the fruitful effects of recombination. As observed in natural ecosystems, higher abiotic diversity leads to higher biotic diversity. [ABSTRACT FROM AUTHOR]

We investigate the evolutionary processes behind the development and optimization of multiple threads of execution in digital organisms using the avida platform, a software package that implements Darwinian evolution on populations of self-replicating computer programs. The system is seeded with a linearly executed ancestor capable only of reproducing its own genome, whereas its underlying language has the capacity for multiple threads of execution (i.e., simultaneous expression of sections of the genome.) We witness the evolution to multithreaded organisms and track the development of distinct expression patterns. Additionally, we examine both the evolvability of multithreaded organisms and the level of thread differentiation as a function of environmental complexity, and find that differentiation is more pronounced in complex environments. [ABSTRACT FROM AUTHOR]

The concept of biodiversity has received rapidly increasing interest in the biosciences during the last decade. Yet, it is unclear and disputed how biodiversity should be characterised and measured. We compared several biodiversity measures by applying them to data retrieved from the LindEvol-GA model of evolution. A series of LindEvol-GA runs with mutation ranges ranging from zero (producing no diversity) to one (producing maximal, but biologically meaningless, diversity) was analyzed with the measures to be compared. At intermediate mutation rates, biologically meaningful diversity can emerge. We show that biodiversity measures can be classified according to the way in which they respond to these various types of diversity, and we discuss some implications of our observation for the design, choice, and application of biodiversity measures. [ABSTRACT FROM AUTHOR]

We use a new general-purpose model of neutral evolution of genotypes to make quantitative comparisons of diversity and adaptive evolutionary activity as a function of mutation rate among two versions of Packard's Bugs model and their neutral shadows. Comparing diversity and evolutionary activity of all these models across the mutation rate spectrum shows that the generic neutral model may have broad applicability in discovering quantitative laws involving adaptive evolutionary activity in different evolving systems. [ABSTRACT FROM AUTHOR]

Recently, Adami and coworkers have been able to measure the information content of digital organisms living in their Avida artificial life system. They show that over time, the organisms behave like Maxwell's demon, accreting information (or complexity) as they evolve. In Avida the organisms don't interact with each other, merely reproduce at a particular rate (their fitness), and attempt to evaluate an externally given arithmetic function in order win bonus fitness points. Measuring the information content of a digital organism is essentially a process of counting the number of genotypes that give rise to the same phenotype. Whilst Avidan organisms have a particularly simple phenotype, Tierran organisms interact with each other, giving rise to an ecology of phenotypes. In this paper, I discuss techniques for comparing pairs of Tierran organisms to determine if they are phenotypically equivalent. I then discuss a method for computing an estimate of the number of phenotypically equivalent genotypes that is more accurate than the "hot site" estimate used by Adami's group. Finally, I report on an experimental analysis of a Tierra run. [ABSTRACT FROM AUTHOR]

Von Neumann's architecture for self-reproducing, evolvable machines is described. Prom this starting point, a number of issues relating to self-reproduction and evolution are discussed. A summary is given of various arguments which have been put forward regarding the superiority of genetic reproduction over self-inspection methods. It is argued that programs in artificial life platforms such as Tierra reproduce genetically rather than by self-inspection (as has previously been claimed). However, the distinction is blurred because significant parts of the reproduction process in Tierran programs are implicitly encoded in the Tierran operating system. The desirable features of a structure suitable for acting as a seed for an open-ended evolutionary process are discussed. It is found that the properties of such a structure are somewhat different to those of programs in Tierra-like platforms. These analyses suggest ways in which the evolvability of individuals in artificial life platforms may be improved, and also point to a number of open questions. [ABSTRACT FROM AUTHOR]

A gene expression system ra-BDD (n-output Binary Decision Diagram) was proposed in order to investigate co-evolution [5]. Although the system is suitable for behavior models of agents, it does not include crossover. This paper proposes a crossover operation using Bryant's Apply operation [2]. The operation makes an n-BDD probabilistically inherit two functions expressed by two n-BDDs. In an experiment the proposed method had more than 40% high fitness than the conventional method. Moreover, in another environment where carnivores and herbivores are co-evolved, we have seen a food chain relation. [ABSTRACT FROM AUTHOR]

Isaacs' treatise on differential games was a break-through for the analysis of the pursuit-and-evasion (PE) domain within the context of strategies representable by differential equations. Current experimental work in Artificial Life steps outside of the formalism of differential games, but the formalism it steps into is yet to be identified. We introduce a formulation of PE that allows a formalism to be developed. Our game minimizes kinematic factors and instead emphasizes the informational aspect of the domain. We use information-theoretic tools to describe agent behavior and implement a pursuit strategy based on statistical decision making; evaders evolved against this pursuit strategy exhibit a wide range of sophisticated behavior that can be quantitatively described. Agent performance is related to these quantifiables. [ABSTRACT FROM AUTHOR]

A version of the standard genetic algorithm, in which the mutation rate is allowed to evolve freely, is applied across a set of optimisation problems. The resulting dynamics confirm the hypothesis that mutation rate, when allowed to evolve, will do so partly as a function of altitude in the fitness landscape. Further, it is demonstrated that this fact can be exploited in order to improve efficiency of the genetic algorithm when applied to a particular class of optimisation problem. Specifically, significant efficiency gains are established in those problems in which the fitness function is not stationary over time. [ABSTRACT FROM AUTHOR]

Whilst the rate at which mutations occur in artificial evolutionary systems has received considerable attention, there has been little analysis of the mutation operators themselves. Here attention is drawn to the possibility that inherent biases within such operators might arte-factually affect the direction of evolutionary change. Biases associated with several mutation operators are detailed and attempts to alleviate them are discussed. Natural evolution is then shown to be subject to analogous mutation "biases". These tendencies are explicable in terms of (i) selection pressure for low mutation rates, and (ii) selection pressure to avoid parenting non-viable offspring. It is concluded that attempts to eradicate mutation biases from artificial evolutionary systems may lead to evolutionary dynamics that are more unnatural, rather than less. Only through increased awareness of the character of mutation biases, and analyses of our models' sensitivity to them, can we guard against artefactual results. [ABSTRACT FROM AUTHOR]

The error threshold — a notion from molecular evolution — is the critical mutation rate beyond which structures obtained by the evolutionary process are destroyed more frequently than selection can reproduce them. We argue that this notion is closely related to the more familiar notion of optimal mutation rates in Evolutionary Algorithms (EAs). This correspondence has been intuitively perceived before ([9], [11]). However, no previous study, to our knowledge, has been aimed at explicitly testing the hypothesis of such a relationship. Here we propose a methodology for doing so. Results on a restricted range of fitness landscapes suggest that these two notions are indeed correlated. There is not, however, a critically precise optimal mutation rate but rather a range of values producing similar near-optimal performance. When recombination is used, both error thresholds and optimal mutation ranges are lower than in the asexual case. This knowledge may have both theoretical relevance in understanding EA behavior, and practical implications for setting optimal values of evolutionary parameters. [ABSTRACT FROM AUTHOR]

We have proposed a theoretical hypothesis called the Programmed Self-Decomposition (PSD) model [1, 2]. The PSD model assumes that a self-decomposition mechanism isprogrammed in each cell of all living organisms on earth, and that thismechanism contributes to the substantial and spatial restoration of the ecosystem. In this article, we would like to introduce the overview of the hypothesis and report on computer simulations in which both a mortal virtual life form based on the PSD model and an immortal virtual life form start their lives in the same finite, heterogeneous ecosystem. Results of the simulations suggest that the mortal life forms would not be exterminated in almost all cases, contrary to ordinary expectations. It seems that the effectiveness of "death" cannot be denied. [ABSTRACT FROM AUTHOR]

In this paper, we describe adaptive processes of populations with two distinct mechanisms of evolution, Darwinian and Lamarckian. We use a simple abstract model where neural networks capable of learning are evolved through GAs. Each individual in the populations tries to maximize its life energy by learning certain rules that distinguish between two groups of materials: food and poison. The best-performing individuals are selected to reproduce offspring according to their mechanism of genetic inheritance, which is either Darwinian or Lamarckian, and the offspring conduct lifetime learning in the succeeding generation. In particular, we examine the adaptability of both populations toward a new unknown world, which is given after some evolutionary steps have taken place under the original world. As the main result, we show that only Darwinian populations can adapt to the new world. [ABSTRACT FROM AUTHOR]

Hinton and Nowlan have demonstrated a model of how lifetime plasticity can guide evolution. They show how acquired traits change the shape of the reward landscape in which subsequent genetic variation takes place, and in so doing encourage the discovery of equivalent heritable traits. This enables the seemingly Lamarkian inheritance of acquired characteristics without the direct transfer of information from the phenotype to the genotype. This paper draws direct inspiration from their work to illustrate a different phenomenon. We demonstrate how the formation of symbiotic relationships in an ecosystem can guide the course of subsequent genetic variation. This phenomenon can be described as two phases: First, symbiotic groups find solutions where individual organisms cannot, simply because lifetime interaction produces new combinations of abilities more rapidly than the relatively slow genetic variation of individuals. Second, these symbiotic groups subsequently change the shape of the reward landscape for evolution, providing a gradient that guides genetic variation to the same solution. Ultimately, an individual organism exhibits the capabilities formerly exhibited by the group. This process enables the combination of characteristics from organisms of distinct species without direct transfer of genetic information. [ABSTRACT FROM AUTHOR]

I will be discussing the plausibility of Mark Bedau's supple adaptation view of life. After presenting the essential aspects of the view, I introduce a hypothetical system of romance novels and romance novel authors, which captures those aspects of a system which are necessary for life under Bedau's view. Using this romance novel system as a starting point, I discuss other intuitively non-living systems, like economies and scientific literature, which would be considered alive by Bedau's standards. Finally, I present one objection to Bedau's view and suggest how his view might change in order to become more widely accepted. [ABSTRACT FROM AUTHOR]

The field of artificial life (Alife) is replete with documented instances of emergence, though debate still persists as to the meaning of this term. In the absence of a formal definition, researchers in the field would be well served by adopting an emergence certification mark which would garner approval from the Alife community. We propose an emergence test, consisting of three criteria—design, observation, and surprise—for conferring the emergence label. [ABSTRACT FROM AUTHOR]