One of the challenges of open-ended learning in robots is the need to autonomously discover goals and learn skills to achieve them. However, when in lifelong learning settings, it is always desirable to generate sub-goals with their associated skills, without relying on explicit reward, as steppingstones to a goal. This allows sub-goals and skills to be reused to facilitate achieving other goals. This work proposes a two-pronged approach for sub-goal generation to address this challenge: a top-down approach, where sub-goals are hierarchically derived from general goals using intrinsic motivations to discover them, and a bottom-up approach, where sub-goal chains emerge from making latent relationships between goals and perceptual classes that were previously learned in different domains explicit. These methods help the robot to autonomously generate and chain sub-goals as a way to achieve more general goals. Additionally, they create more abstract representations of goals, helping to reduce sub-goal duplication and make the learning of skills more efficient. Implemented within an existing cognitive architecture for lifelong open-ended learning and tested with a real robot, our approach enhances the robot’s ability to discover and achieve goals, generate sub-goals in an efficient manner, generalize learned skills, and operate in dynamic and unknown environments without explicit intermediate rewards.

Autonomously acquiring grounded information from on-line interaction in continuous and dynamic domains unknown at design time to allow for abstraction and high-level reasoning is still a challenging problem in robotics. The main issues are how to create the appropriate categories from non-directed interaction with the world, to do so in a way that allows for composition so that future learning is facilitated, and to meet the adaptability requirements imposed by ever changing domains. Thus, in this paper we present an approach based on the generation of task-based dynamic perceptual equivalence classes that are constantly updated and adapted during the life of the robot. To provide a real example, the approach was implemented using three different types of algorithms within the e-MDB cognitive architecture on a real robot. We present results of the robot interacting with different domains and demonstrate the adaptability of the approach through a series of experiments in which the domain is changed arbitrarily. We compare the performance of the three algorithms that were implemented and discuss the consequences of this approach and its possible uses within cognitive architectures. These autonomously generated perceptual classes and their appropriate representation within cognitive architectures constitute a path towards more abstract (and even symbolic) processing starting from grounded emergent low-level components.

In human beings, the joint development of the body and cognitive system has been shown to facilitate the acquisition of new skills and abilities. In the literature, these natural principles have been applied to robotics with mixed results and different authors have suggested several hypotheses to explain them. One of the most popular hypotheses states that morphological development improves learning by increasing exploration of the solution space, avoiding stagnation in local optima. In this article, we are going to study the influence of growth-based morphological development and its nuances as a tool to improve the exploration of the solution space. We will perform a series of experiments over two different robot morphologies which learn to walk. Furthermore, we will compare these results to another optimization strategy that has been shown to be useful to favor exploration in learning algorithms: the application of noise during learning. Finally, to check if the increased exploration hypothesis holds, we visualize the genotypic space during learning considering the different optimization strategies by using the Search Trajectory Network representation. The results indicate that noise and growth increase exploration, but only growth guides the search towards good solutions.

The current work is focused on providing robotic systems with mechanisms that support lifelong open-ended learning, with the aim of increasing their real autonomy level. In this general scope, robots must discover their goals and learn the skills to achieve them in a priori unknown domains and tasks. Moreover, they must do it in way that allows this knowledge to be reused later to face new situations properly. To advance in this challenging field, this paper describes a specific motivation-based knowledge reuse strategy, together with the contextual processing carried out, within the e-MDB cognitive architecture. This strategy supports the reuse and adaptation of knowledge acquired in previously seen domains to new ones. It has been validated in a real-world experiment with the Baxter robot, which is analyzed and discussed here, that addresses open-ended interaction in a sequence of domains related to object manipulation.

In living beings, the natural development of the body has been shown to facilitate learning. The application of these natural developmental principles in robotics have been considered in different robotic morphologies and scenarios, leading to mixed results. Development was found to be beneficial for learning in some instances, but also irrelevant or detrimental in others. This mix of results and scenarios has allowed researchers to extract some notions about the conditions that must be fulfilled or set to apply morphological development successfully. Notions that we have organized to set a series of design conditions to successfully apply morphological development. Thus, in this article, we are going to focus on the study of one of them that has been frequently addressed by researchers in their studies in very general terms. It can be described as the need to achieve a suitable synergy among the different components involved in the development and learning process: morphological development strategy, controller, task, and learning algorithm. In particular, we have concentrated on empirically determining the influence of five developmental strategies, implemented in different ways, applied at different speeds and deployed in different orders and combinations, over the problem of a NAO robot controlled by an artificial neural network obtained through a neuroevolutionary algorithm learning a bipedal walking task. The results obtained permit providing a more detailed description of what a suitable synergy implies and how it can be utilized to design more successful morphological developmental processes to improve robot learning.

In human beings, the natural development of the body has been shown to facilitate learning. This approach has been applied in robotic learning with different results, being an advantage under some conditions and tasks. While it is still not well understood under what conditions morphological development helps to learn, several authors have proposed some high-level notions about when it could be interesting to apply it. In our previous work, we have used these notions with the objective of designing a morphological development strategy that facilitates learning in a bipedal locomotion task with an Artificial Neural Network (ANN) controlled robot. In this paper, we aim to go beyond the qualitative design principles previously used and support such considerations with an empirical quantitative study. An analysis of the learning results and how they are related to the design conditions that were established is carried out based on the evolution of the fitness landscape for each developmental stage. The long-term objective is to develop morphology-agnostic optimization strategies for morphological development, which would reduce the number of samples required and, thus, the computational cost, of learning in ANN-controlled robots.

The frontier in robot autonomy is currently Lifelong Open-Ended Autonomy (LOLA). Within these settings, a robot must be able to operate and learn in domains that are unknown at design time as well as reuse knowledge learnt in one domain to facilitate learning in others throughout its lifetime. Achieving LOLA goes beyond learning specific algorithms and puts us squarely in the realm of cognitive architectures; however, most cognitive architectures were not built to address the LOLA problem, and thus, lack components and capabilities that would be required for it. In fact, even though there is a growing literature on learning representations, especially in the framework of reinforcement and deep learning, hardly any cognitive architecture considers the issue of autonomously learning representations, which is a crucial problem to be able to efficiently learn and abstract information when seeking LOLA. This paper provides a vision of the general requirements in terms of learning knowledge representations within cognitive architectures geared towards LOLA and addresses a specific problem in this context: the problem of learning representations that facilitate obtaining world and utility models and deciding on actions in situations where multiple goals can be activated. The work is carried out in the framework of the development of the e-MDB cognitive architecture for real robots operating autonomously in real domains.

In nature, the physical development of the body that takes place in parallel to the cognitive development of the individual has been shown to facilitate learning. This opens up the question of whether the same principles could be applied to robots in order to accelerate the learning of controllers and, if so, how to apply them effectively. In this line, several authors have run experiments, usually quite complex and heterogeneous, with different levels of success. In some cases, morphological development seemed to provide an advantage and in others it was clearly irrelevant or even detrimental. Basically, morphological development seems to provide an advantage only under some specific conditions, which cannot be identified before running an experiment. This is due the fact that there is still no agreement on the underlying mechanisms that lead to success or on how to design morphological development processes for specific problems. In this paper, we address this issue through the execution of different experiments over a simple, replicable, and straightforward experimental setup that makes use of different neural network controlled walkers together with a morphological development strategy based on growth. The morphological development processes in these experiments are analyzed both in terms of the results obtained by the different walkers and in terms of how their fitness landscapes change as the morphologies develop. By comparing experiments where morphological development improves learning and where it does not, a series of initial insights have been extracted on how to design morphological development processes.

This paper seeks to explore the effect and possibilities of autonomously balancing drives in a motivational architecture aimed at open-ended learning robots. These types of robots are very useful in unconstrained human robot interaction settings or when uncontrolled dynamic scenarios that are unknown at design time must be addressed. Designing a robot under these conditions implies that it must be endowed with some primary operational purpose and some additional self-preservation objectives whose fulfillment depend on the characteristics of the particular domain it is facing each moment in time. Domains that are not known beforehand and for which no a priori goal or skill structure can be designed in. Thus, an approach to the design and engineering of motivational structures to endow robots with specific purposes is proposed and tested here. We concentrate on the drive structure of a motivational system and the effects of its autonomous adaptation to changing circumstances. To provide for this adaptation, a simple evolutionary strategy is defined for the autonomous regulation of multiple drives seeking to optimize long-term operation. The proposal is tested on a Baxter robot performing an industrial task and the results confirm the potential of autonomous dynamic drive balancing as a tool in open-ended settings.

Morphological development is part of the way any human or animal learns. The learning processes starts with the morphology at birth and progresses through changing morphologies until adulthood is reached. Biologically, this seems to facilitate learning and make it more robust. However, when this approach is transferred to robotic systems, the results found in the literature are inconsistent: morphological development does not provide a learning advantage in every case. In fact, it can lead to poorer results than when learning with a fixed morphology. In this paper we analyze some of the issues involved by means of a simple, but very informative experiment in quadruped walking. From the results obtained an initial series of insights on when and under what conditions to apply morphological development for learning are presented.

Motivation is quite an important topic when addressing continual open-ended learning processes in autonomous robots. The three main issues that need to be considered are, firstly, how does a designer define what the robot strives for in a manner that is independent from any particular domain it may find itself in. Secondly, once that robot is in a domain, how does it go about finding and relating goals in that particular domain on its own. Finally, the third issue is, once a goal is found, how does a robot establish a representation, usually in the form of a Value Function, that will allow it to exploit that goal. This paper deals with the third issue in the framework of the motivational engine we have designed for cognitive architectures. It addresses the problem of efficiently and appropriately learning complex Value Functions starting from intrinsically motivated traces of valuated robot actions that are often ambiguous and multivalued. To this end, a developmental learning mechanism is proposed that relies on the concurrent application of a real time ANN learning procedure over the traces of the valuated robot actions, and a simpler sensor correlation-based approach to allow for the production of better configured data traces for the learning process. The mechanism is analyzed and discussed over an experiment considering a real Baxter robot.

Artificial intelligence and all its supporting tools, e.g. machine and deep learning in computational intelligence-based systems, are rebuilding our society (economy, education, life-style, etc.) and promising a new era for the social welfare state. In this paper we summarize recent advances in data science and artificial intelligence within the interplay between natural and artificial computation. A review of recent works published in the latter field and the state the art are summarized in a comprehensive and self-contained way to provide a baseline framework for the international community in artificial intelligence. Moreover, this paper aims to provide a complete analysis and some relevant discussions of the current trends and insights within several theoretical and application fields covered in the essay, from theoretical models in artificial intelligence and machine learning to the most prospective applications in robotics, neuroscience, brain computer interfaces, medicine and society, in general.

The modeling of utility is an important problem in many fields, including reinforcement learning. However, when considering a developmental approach to open-ended learning a new aspect arises. In these settings, the efficiency of the modeling process becomes a key aspect, as these processes usually take place in real time and, to increase survivability, it is necessary for the robot to be able to produce utility models as fast as possible. In this paper, we address this issue by proposing a modulation-based approach to the adaptation of the robot's experience, in the form of previously obtained ANN based utility models, to new situations. These previous utility models are perceptually recalled from a Long-Term Memory and combined to produce an initial guess to the new utility model. After this, modulatory structures are created that lead to the fine adaptation of these initial guesses to the real utility model of the new situation. Some initial results of experiments using a real robot are presented to clarify the approach. Specifically, three realistic problems that a Baxter "cooking robot" must solve are faced with this modulating approach. With them, it is clearly shown the increase in efficiency of the utility model learning in real time.

In the framework of open-ended learning cognitive architectures for robots, this paper deals with the design of a Long-Term Memory (LTM) structure that can accommodate the progressive acquisition of experience-based decision capabilities, or what different authors call "automation" of what is learnt, as a complementary system to more common prospective functions. The LTM proposed here provides for a relational storage of knowledge nuggets given the form of artificial neural networks (ANNs) that is representative of the contexts in which they are relevant in a configural associative structure. It also addresses the problem of continuous perceptual spaces and the task- and context-related generalization or categorization of perceptions in an autonomous manner within the embodied sensorimotor apparatus of the robot. These issues are analyzed and a solution is proposed through the introduction of two new types of knowledge nuggets: P-nodes representing perceptual classes and C-nodes representing contexts. The approach is studied and its performance evaluated through its implementation and application to a real robotic experiment.

This paper is concerned with the problem of perceptual classification in the framework of life-long learning developmental cognitive architectures. Perceptual classification is the process by which autonomous entities organize their, usually continuous, perceptual streams into classes of perceptions that are relevant to the different contexts in which they find themselves. In particular, here we describe an approach based on context related generalization or categorization of perceptions in an autonomous manner within embodied systems. This approach involves the introduction of a new type of knowledge nuggets, Pnodes, within the long term memory structure of a cognitive architecture. Taking inspiration from the hippocampus-cortex relationships in real brains, P-nodes are initially described by means of a set of representative perceptual points which are subsequently generalized in a cortex like neural representation. This approach is tested in a series of experiments on a real robot.

This paper describes a re-descriptive approach to the efficient acquisition of ever higher level and more precise utility models within the motivational system (MotivEn) of a cognitive architecture. The approach is based on a two-step process whereby, as a first step, simple imprecise sensor correlation related utility models are obtained from the interaction traces of the robot. These utility models allow the robot to increase the frequency of achieving goals, and thus, provide lots of traces that can be used to try to train precise value functions implemented as artificial neural networks. The approach is tested experimentally on a real robotic setup that involves the coordination of two robots.

Abstract This work addresses the problem of denoising image sequences through an approach that makes use of spatio‐temporal cellular automata‐based filtering. The algorithm is called st‐CAF and one of its key aspects is that the resulting cellular automata contemplate a spatio‐temporal neighbourhood when processing each pixel of the sequence. Additionally, the way the rule sets for the cellular automata are obtained, through evolutionary means, is also relevant, as it allows a good adaptation to any type of image and noise through the appropriate training set. This results in a great advantage over more traditional single frame denoising techniques presented in the literature or even over their adaptation to sequences. A fact that is made relevant in this paper through the application of the algorithm to different types of noisy images and its comparison to other techniques.

This work describes a novel spatio-temporal cellular automata-based filtering algorithm (st-CAF) intended for performing image sequence denoising processes. The approach presents several advantages over more traditional single frame denoising techniques presented in the literature or even over their adaptation to sequences. Especially the fact that the cellular automaton used is able to contemplate information concerning the type of noise through the use of specific sequences to tune the algorithm, as well as temporal information by means of a spatio-temporal neighborhood when processing each pixel of the sequence. These two elements lead to significant improvements in the results with respect to simple spatial or temporal sets of neighbors.

Embodied Evolution (EE) is an evolutionary strategy based on natural evolution in which the individuals that make up the population are embodied and situated in an environment where they interact in a local, decentralized and asynchronous fashion. It has been successfully applied in collective problems showing its validity to perform on-line evolution both in simulated and real agents. A key feature of EE is that of emergent specialization, that is, this strategy is able to autonomously generate a distribution of individuals into species if that is advantageous in the scenario. This paper goes in the line of studying such feature in more depth, analyzing how the complexity of the task (fitness landscape) and the complexity of the individuals (control system) affect the emergence of specialization. The analysis is carried out using a canonical EE algorithm in a real problem consisting in a collective surveillance task with simulated Micro Aerial Vehicles.

This paper presents an initial integrated approximation to the complete problem of robot motivation in continuous domains in terms of how to adaptively combine intrinsic and extrinsic motivations into an integrated motivational engine, called MotivEn. It allows an autonomous robot to find goals and decompose them into sub-goals that can be chained to facilitate achieving the final goal. MotivEn is based on an evolutionarily learnt value function in continuous domains where exploration and exploitation, as well as its decomposition into sub-value functions, is autonomously achieved.

This paper proposes an intrinsically distributed cellular automata (CA) based approach to address the perennial problem of real time segmentation and classification of high dimensional images, such as remote sensing hyperspectral images. This approach is efficiently implemented on GPUs providing results that improve on the state of the art algorithms presented in the literature. It is based on the evolutionary generation of the CA rule sets under two basic premises: During the segmentation process, the CAs must work over the whole dimensionality of the images without any projection onto lower dimensionalities, and the rule sets that are generated must be adapted to the segmentation level required by the user. The performance of the approach is tested over a benchmark set of well-known hyperspectral images and the results compared to the state of the art in the literature for two implementations, one using a SVM based classification stage and another that considers an ELM based classification stage.

In this paper we argue that to produce good optimization performances, the exploration of the solution space does not need to be carried out in the unorderly fashion most evolutionary algorithms use. Other strategies that seek to minimize the cost involved in successive evaluation processes should be explored. This does not imply a fundamental change on how evolutionary algorithms work, but rather, it brings some structure onto how solution spaces are explored by contemplating decoding cost as one of the elements to be minimized when sampling. The traditional implementations of most evolutionary algorithms assume that any point in the solution space can be evaluated any time and at no cost. However, this is not always the case and often each step of the process only part of the solution space is available for evaluation giving rise to a class of problems we have called Constrained Sampling optimization problems over which evolutionary algorithms are quite inefficient. To address these problems we have proposed a modification of the general strategy of evolutionary algorithms to address these constraints efficiently. Here, we study the effects of this approach when applied to problems that are not constrained, thus modifying the way the solution space is explored. This study is carried out to determine how these modification impact the performance of a set of popular evolutionary algorithms over a representative set of benchmark functions corresponding to fitness landscapes with a variety of characteristics. We show that by restricting the sampling capabilities of most algorithms, the cost of the optimization procedure is reduced for most types of fitness landscapes without affecting their results.

In this paper we describe a hybrid evolutionary-cellular automata based algorithm for the segmentation of multidimensional images, in particular hyperspectral images. This algorithm permits automatically generating the cellular automata transition rule set using as training set a group of appropriately generated synthetic RGB images, which greatly simplifies the process given the lack of adequately labeled hyperspectral images. In addition, different types of high dimensional segmentations can be obtained through the regulation of the parameters of the RGB images in the training set. The algorithm has been tested over synthetic and real hyperspectral images and the segmentation results it produces are very competitive when compared to other approaches found in the literature.

The automatic design of control systems for multi-robot teams that operate in real time is not affordable with traditional evolutionary algorithms mainly due to the huge computational requirements they imply. Embodied Evolution (EE) is an evolutionary paradigm that aims to address this problem through the embodiment of the individuals that make up the population in the physical robots. The interest for this type of evolutionary approach has been increasing steadily, leading to different algorithms and variations adapted to solve very specific practical cases. In a previous work, the authors started the implementation of a standard canonical EE algorithm that captures the more general principles of this paradigm and that can be applied to any distributed optimization problem. This canonical algorithm has been characterized already over a set of theoretical fitness landscapes corresponding to representative examples of the basic casuistry found in collective tasks. The current paper goes one step ahead in this research line, and the canonical algorithm is applied here in a collective navigation task in which a fleet of Micro Aerial Vehicles (MAVs) has to gather red rocks in an indoor scenario. The objective is to confirm that the characterization conclusions are generalizable to a practical case and to show that the canonical algorithm can be configured to operate as a specific algorithm easily.

This work proposes a set of modifications to the Differential Evolution algorithm in order to make it more efficient in solving a particular category of problems, the so called Constrained Sampling problems. In this type of problems, which are usually related to the on-line real-world application of evolution, it is not always straightforward to evaluate the fitness landscapes due to the computational cost it implies or to physical constraints of the specific application. The fact is that the sampling or evaluation of the offspring points within the fitness landscape generally requires a decoding phase that implies physical changes over the parents or elements used for sampling the landscape, whether through some type of physical migration from their locations or through changes in their configurations. Here we propose a series of modifications to the Differential Evolution algorithm in order to improve its efficiency when applied to this type of problems. The approach is compared to a standard DE using some common real-coded benchmark functions and then it is applied to a real constrained sampling problem through a series of real experiments where a set of Unmanned Aerials Vehicles is used to find shipwrecked people.

This paper is concerned with the incorporation of new time processing capacities to the Neuroevolution of Augmenting Topologies (NEAT) algorithm. This algorithm is quite popular within the robotics community for the production of trained neural networks without having to determine a priori their size and topology. However, and even though the algorithm can address temporal processing issues through its capacity of establishing feedback synaptic connections, that is, through recurrences, there are still instances where more precise time processing may go beyond its limits. In order to address these cases, in this paper we describe a new implementation of the NEAT algorithm where trainable synaptic time delays are incorporated into its toolbox. This approach is shown to improve the behavior of neural networks obtained using NEAT in many instances. Here, we provide some of these results using a series of typical complex time processing tasks related to chaotic time series modeling and consider an example of the integration of this new approach within a robotic cognitive architecture.

This work presents a novel spatio-temporal cellular automata-based filtering (STCAF) for image sequence denoising. Most of the methods using cellular automata (CA) for image denoising involve the manual design of the rules that define the behaviour of the automata. This is a complex and not straightforward operation. In order to tackle this problem, this paper proposes to use evolutionary methods to obtain the CA set of rules which produces the best possible denoising under different noise models or/and image sources. This is implemented using a spatio-temporal neighbourhood for each pixel, which significantly improves the results with respect to simple spatio or temporal set of neighbours. The proposed method is tested to reduce the noise in low-dose X-ray image sequences. These data have a severe signal-dependent noise that must be reduced avoiding artifacts while preserving structures of interest for a medical inspection. The proposed method outperforms several state-of-the-art algorithms on both simulated and real sequences.

This paper is concerned with the design and implementation of a new concept of robot to clean the underwater sections of ship hulls without using any magnetic attachment. The use of this type of robots on a regular basis to preserve a clean hull, usually when ships are in port or anchored, will improve the efficiency of the ships and will permit a reduction in the use of chemicals that are harmful to the environment to prevent the growth of marine life on the hull. The main contribution of the robot described in this paper is that it is a completely novel design that through an appropriate morphology solves the problems that arise when moving along hulls, including changing planes, negotiating appendices, portholes, corners, and other elements. It thus provides a basis for completely autonomous operation. The design and implementation of the robot is described and some simulations and tests in real environments are presented.

The performance of any Evolutionary Algorithm (EA) is closely related to the topographical features of the problem fitness landscape it is applied to. It is therefore of paramount importance to determine a set of features that is useful in order to choose an appropriate algorithm for a given problem. This way, the inefficient trial and error stage that most EA users carry out until they find an EA that satisfies their objectives can be reduced. In fact, as this, usually lengthy, trial and error stage is generally carried out in an ad hoc manner, the information the user gleans from the performance of the algorithms chosen and their particular parameter sets, or lack thereof, can be very misleading or plain useless. Thus, in previous work, we analyze a set of features in synthetic fitness landscapes that can be used in order to characterize problems and relate them to the performance of EAs. The objective is to define a mechanism to reduce the trial and error stage when choosing the correct EA and, at the same time, provide more in depth knowledge on the nature of the problem. Here, in order to highlight the usefulness of the approach, this analysis is extended to real world application landscapes by means of the characterization of a horizontal axis wind turbine (HAWT) design problem, showing the relevance of the pre-processing stage in the selection of the most appropriate EA to solve it.

This paper deals with the problem of segmenting or, more properly, finding edges in multidimensional images, in particular, hyperspectral images. The approach followed is based on the use of cellular automata (CA) and their emergent behavior in order to achieve this objective. Using cellular automata for finding edges in hyperspectral images is not new, but most current approaches to this problem involve hand designing the rules for the automata. On the other hand, many authors just use extensions of one-dimensional edge detection methods to multidimensional images, thus averaging out the spectral information present. Here, we consider the application of evolutionary methods to produce the CA rule sets that obtain the best possible edge detection properties under different circumstances and using spectral based approaches. The procedure has been tested over synthetic and real hyperspectral images and the results obtained have been compared to those produced using the hyper-Sobel and Hyper-Prewitt operators, which are standard edge detection methods for gray-level images that have been extended by some authors to the multidimensional domain.

The Body Condition Score (BCS) is considered a critical value for dairy farms, since its observation can be used to optimize milk production. Usually, the BCS is calculated by human experts after visual inspection in a time-consuming and subjective process. There are already some papers where this process is almost automated using image processing on some kinds of pictures and, in this work, the first steps towards a fully automated method based on pictures taken with common photographic cameras are described. Active Shape Models (ASM) are used to obtain a set of features that describe the back shape of cows and those features feed a classifier that computes the BCS. We show that the BCS can be estimated using only a set of angles from the back view with an error similar to that calculated between scores of two experts. To obtain those angles automatically is the hardest step in this process, but we have already achieved reasonable results on that point too.

The problem of segmenting multidimensional images, in particular hyperspectral images, is still an open subject. The main issue is related to preserving the multidimensional character of the signals throughout the segmentation process avoiding an early projection onto a 2D plane with the consequent loss of the wealth of information these images provide. The approach followed here is based on the use of cellular automata (CA) and their emergent behavior over the hyperspectral cube in order to achieve this objective. Using cellular automata for segmentation in hyperspectral images is not new, but most approaches to this problem involve hand designing the rules for the automata. Additionally, most references found are just extensions of one or three-dimensional methods to multidimensional images, and, as a consequence, average out the spectral information present. The main contributions of this paper is the study of the application of evolutionary methods to produce the CA rule sets that result in the best possible segmentation properties under different circumstances without resorting to any form of projection until the information is presented to the user. The procedure has been tested over synthetic and real hyperspectral images.

This paper is devoted to a detailed presentation of the current state of the Multilevel Darwinist Brain (MDB) cognitive architecture for lifelong learning in real robots. This architecture follows the cognitive developmental robotics approach and it is based on concepts like embodiment, open-ended lifelong learning, autonomous knowledge acquisition or adaptive behaviors and motivations. In addition, this version of the MDB architecture incorporates several improvements related with more practical issues, which are the result of the experience gained through several experiments with real robots in the last few years. The MDB uses evolutionary algorithms in the knowledge acquisition process, which implies the need of paying attention to the efficiency of the computational implementation. Here, we first describe the cognitive model on which the basic operation of the architecture is based and, secondly, we detail the main aspects and working of the current version of the MDB. Finally, we have designed a very simple but illustrative real robot lifelong learning example, where we can show how to set up an experiment using the MDB. Hence, with this simple example we show the successful behavior of the MDB cognitive developmental robotics principles.

This paper describes an artificial neural network based system for classifying the contents of hyperspectral images that is able to automatically reduce the dimensionality of the data provided by the hyperspectrometers without compromising their efficacy. The data reduction is achieved through the adaptation of the window size and the number of parameters that make up the description of the spectral signatures within the window as training progresses. Following this approach, a user just needs to specify the minimum resolution desired on the output or category image and the level of discrimination among categories, and the system will try to meet these requirements by modifying during training the size and number of inputs to the network. When it is not possible to comply with both requirements, the system will provide a compromise solution that minimizes the global discrimination error, which takes into account the spatial discrimination and the discrimination among classes.

In this paper we discuss the relevance of performing a population dynamics analysis to improve the results obtained using agent-based artificial life systems for optimization. The present study derives from our work trying to solve engineering optimization problems using a distributed approach based on agent's interactions. We have realized that a simple analysis of the population dynamics can show the relevance of some variables and energy exchange rates in the stability of the system. The results obtained can be used to control the equilibrium points and/or avoid non-convergence (population extinctions) by changing the initial conditions or the parameters of the energetic model used in the system. To illustrate the results of such population dynamics analysis, a practical example based on a routing algorithm is presented.

This paper deals with the design and construction of a climbing robot for performing grit blasting operations in shipyards. The robot is based on a double sliding platform that uses permanent magnets for attachment. It is lightweight and compact and can move up and along the shipside with any inclination while grit blasting the surface to pre-specified surface quality levels. It can also rotate to compensate for hull curvature and to avoid obstacles while performing its task. The blasting operation is modulated by a vision based quality control system that is used by the mission control system to adapt the blasting parameters in order to attain the desired quality levels while maximizing the surface area the robot strips per unit time.

This paper deals with the problem of comparing and testing evolutionary algorithms, that is, the benchmarking problem, from an analysis point of view. A practical study of the application domain of four representative evolutionary algorithms is carried out using a relevant set of real-parameter function optimization benchmarks. The four selected algorithms are the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and the Differential Evolution (DE), due to their successful results in recent studies, a Genetic Algorithm with real parameter operators, used here as a reference approach because it is probably the most familiar to researchers, and the Macroevolutionary algorithm (MA), which is not widely known but it shows a very remarkable behavior in some problems. The algorithms have been compared running several tests over the benchmark function set to analyze their capabilities from a practical point of view, in other words, in terms of their usability. The characterization of the algorithms is based on accuracy, stability and time consumption parameters thus establishing their operational scope and the type of optimization problems they are more suitable for.

Traditional schedulers for high performance computing (HPC) systems are, nowadays, powerful but hard to configure, with a large number of parameters. They are not very flexible and they ignore two things: many users don't know or specify the resources needed by their jobs, and the same system performance is not perceived equally by every user. The work presented here is focused on the task of improving scheduling by addressing these problems through a system based on three key components: genetic algorithms, user behavior models and user satisfaction models. Thus, a genetic algorithm tries to find, with no human intervention, the best order for the execution of jobs using an automatically obtained behavior model for each user that predicts the real amount of resources needed, and a fitness function that takes into account the concept of user satisfaction (extracted from a user satisfaction model) in addition to classical parameters such as makespan or waiting times. The complete scheduling system is described as well as its integration in the Sun Grid Engine (SGE). Some experiments are carried out to compare its behavior to that of the SGE, including the effects of using different parameters in the fitness function in order to consider different needs or policies of the HPC center. Some comments are provided on how user satisfaction models affect scheduling.

In this article we describe the use of Artificial Neural Networks (ANN) with synaptic time delays between the nodes as a means to increase the capabilities of the usual control modules used in behavior based robotics. This inclusion allows the controllers to manage explicit temporal information in different levels. In the sensing level it permits the use of virtual sensors that improve the precision of the information provided by sensors through a temporal correlation of their values. In the actuation level we use the network with an infrasensorized robot in a problem that requires active sensing, where the control and actuation mechanisms are coordinated in order to obtain a better sensorial image of the environment by means of a spatio-temporal representation of a perception sequence. The decision of the appropriate delays is automated through learning and evolution.

In the context of visual systems for robots, we have made use of a high order gaussian synapses network and the Gaussian Synapses Backpropagation Algorithm (GSBP) for the implementation of the detectors that constitute one part of the whole visual architecture. These detectors are trained to be sensitive to spatial patterns that are relevant for the decisions the robot must perform during its operation in an environment. The inclusion of gaussian functions in the synapses of the network allows the network to select the appropriate spatial information and filter out all that is irrelevant according to the training it has received. In this paper we will show how these networks are easily trained to ignore backgrounds. In addition, with a very simple training set and an appropriate input selection strategy, the networks detect objects independently of size and position. These systems, coupled with an attention mechanism result in a very efficient visual information processor.

The aim of this paper is to endow a well-known structure for processing time-dependent information, synaptic delay-based ANN's, with a reliable and easy to implement algorithm suitable for training temporal decision processes. In fact, we extend the backpropagation algorithm to discrete-time feedforward networks that include adaptable internal time delays in the synapses. The structure of the network is similar to the one presented by [1], that is, in addition to the weights modeling the transmission capabilities of the synaptic connections, we model their length by means of a parameter that indicates the delay a discrete-event suffers when going from Zthe origin neuron to the target neuron through a synaptic connection. Like the weights, these delays are also trainable, and a training algorithm can be derived that is almost as simple as the backpropagation algorithm, and which is really an extension of it. We present examples of the application of these networks and algorithm to the prediction of time series and to the recognition of patterns in electrocardiographic signals. In the first case, we employ the temporal reasoning characteristics of these networks for the prediction of future values in a benchmark example of a time series: the one governed by the Mackey-Glass chaotic equation. In the second case, we provide a real life example. The problem consists in identifying different types of beats through two levels of temporal processing, one relating the morphological features which make up the beat in time and another one that relates the positions of beats in time, that is, considers rhythm characteristics of the ECG signal. In order to do this, the network receives the signal sequentially, no windowing, segmentation, or thresholding are applied.