Aleksandar Jevtić received his B.Sc. and M.Sc. degrees in Electrical Engineering from the University of Belgrade, Serbia, in 2005. He received his M.Sc. and Ph.D. ("Cum Laude", with European Ph.D. mention) degrees in Computer Science from the Technical University of Madrid (UPM), Spain, in 2007 and 2011, respectively. Presently, he works as a Marie Curie Fellowship postdoctoral researcher at Robosoft, France. His research interests include human-robot interaction, swarm intelligence, swarm robotics, multi-agent systems, data mining, image processing and analysis, and financial markets.
Abstract: In this paper, we propose the distributed bees algorithm (DBA) for task allocation in a swarm of robots. In the proposed scenario, task allocation consists in assigning the robots to the found targets in a 2-D arena. The expected distribution is obtained from the targets' qualities that are represented as scalar values. Decision-making mechanism is distributed and robots autonomously choose their assignments taking into account targets' qualities and distances. We tested the scalability of the proposed DBA algorithm in terms of number of robots and number of targets. For that, the experiments were performed in the simulator for various sets of parameters, including number of robots, number of targets, and targets' utilities. Control parameters inherent to DBA were tuned to test how they affect the final robot distribution. The simulation results show that by increasing the robot swarm size, the distribution error decreased.
Abstract: Swarms of robots can use their sensing abilities to explore unknown environments and deploy on sites of interest. In this task, a large number of robots is more effective than a single unit because of their ability to quickly cover the area. However, the coordination of large teams of robots is not an easy problem, especially when the resources for the deployment are limited. In this paper, the Distributed Bees Algorithm (DBA), previously proposed by the authors, is optimized and applied to distributed target allocation in swarms of robots. Improved target allocation in terms of deployment cost efficiency is achieved through optimization of the DBAâs control parameters by means of a Genetic Algorithm. Experimental results show that with the optimized set of parameters, the deployment cost measured as the average distance traveled by the robots is reduced. The cost-efficient deployment is in some cases achieved at the expense of increased robotsâ distribution error. Nevertheless, the proposed approach allows the swarm to adapt to the operating conditions when available resources are scarce.
Abstract: Metaplasticity property of biological synapses is interpreted in this paper as the concept of placing greater emphasis on training patterns that are less frequent. A novel implementation is proposed in which, during the network learning phase, a priority is given to weight updating of less frequent activations over the more frequent ones. Modeling this interpretation in the training phase, the hypothesis of an improved training is tested on the Multilayer Perceptron type network with Backpropagation training. The results obtained for the chosen application show a much more efficient training, while at least maintaining the Multilayer Perceptron performance.
Abstract: Edge detection is a pre-processing step in applications of computer and robot vision. The aim is to extract the features or objects of interest from a digital image and obtain information about the environment. Two edge-detection methods inspired by ants foraging behavior are proposed. The ants in nature use pheromone to mark the path to the food source. In digital images, the ant-like agents move through the discrete space of an image and search for edge pixels. The first proposed method extracts the edges from a grayscale image. The second method finds the missing segments of the broken edges and it can be applied as a complementary tool to any edge detector. Finally, a study on the adaptability of the first edge detector was performed on a set of grayscale images that were used to create a dynamically changing environment. A set of simulations was performed to qualitatively and quantitatively evaluate the proposed methods.
Abstract: This chapter introduces a swarm intelligence-inspired approach for target allocation in large teams of autonomous robots. For this purpose, the Distributed Bees Algorithm (DBA) was proposed and developed by the authors. The algorithm allows decentralized decision-making by the robots based on the locally available information, which is an inherent feature of animal swarms in nature. The algorithm's performance was validated on physical robots. Moreover, a swarm simulator was developed to test the scalability of larger swarms in terms of number of robots and number of targets in the robot arena. Finally, improved target allocation in terms of deployment cost efficiency, measured as the average distance traveled by the robots, was achieved through optimization of the DBA's control parameters by means of a genetic algorithm.
Abstract: For a human-robot interaction to take place, a robot needs to perceive humans. The space where a robot can perceive humans is restrained by the limitations of robotâs sensors. These restrictions can be circumvented by the use of external sensors, like in intelligent environments; otherwise humans have to ensure that they can be perceived. With the robotic platform presented here, the roles are reversed and the robot autonomously ensures that the human is within the area perceived by the robot. This is achieved by a combination of hardware and algorithms capable of autonomously tracking the person, estimating their position and following them, while recognizing their gestures and navigating through environment.
Abstract: Person following by a mobile autonomous robot includes two tasks, person tracking and safe robot navigation. Two person-following algorithms that use depth images from a Microsoft Kinect sensor for person tracking are proposed. The first one, the path-following algorithm, reproduces the path of the person in the environment. The second one, the adaptive algorithm, uses in addition a laser range finder for localization and dynamically generates the robotâs path inside
a pre-mapped environment, taking into account the obstacles locations. The Kinect was mounted on a pan-tilt mechanism to allow continuous person tracking while the robot followed the generated path. The two algorithms were tested and their performance compared in a series of trials where the robot had to follow a person walking in an environment with obstacles. With both algorithms the robot could perform continuous person tracking when the obstacles were lower than the height of the camera mount. With the adaptive algorithm the distance traveled by the robot was 29.6% shorter than with the path-following algorithm; however the path-following algorithm does not require a pre-build map of the environment.
Abstract: Industrial applications of computer vision sometimes require detection of atypical objects that occur as small groups of pixels in digital images. These objects are difficult to single out because they are small and randomly distributed. In this work we propose an image segmentation method using the novel Ant System-based Clustering Algorithm (ASCA). ASCA models the foraging behaviour of ants, which move through the data space searching for high data-density regions, and leave pheromone trails on their path. The pheromone map is used to identify the exact number of clusters, and assign the pixels to these clusters using the pheromone gradient. We applied ASCA to detection of microcalcifications in digital mammograms and compared its performance with state-of-the-art clustering algorithms such as 1D Self-Organizing Map, k-Means, Fuzzy c-Means and Possibilistic Fuzzy c-Means. The main advantage of ASCA is that the number of clusters needs not to be known a priori. The experimental results show that ASCA is more efficient than the other algorithms in detecting small clusters of atypical data.
Abstract: Breast cancer is one of the leading causes to women mortality in the world and early detection is an important means to reduce the mortality rate. The presence of microcalcifications clusters has been considered as a very important indicator of malignant types of breast cancer and its detection is important to prevent and treat the disease. This paper presents an alternative and effective approach in order to detect microcalcifications clusters in digitized mammograms based on the synergy of the image processing, pattern recognition and artificial intelligence. The mathematical morphology is an image processing technique used for the purpose of image enhancement. A k-means algorithm is used to cluster the data based on the features vectors and finally an artificial neural network-based classifier is applied and the classification performance is evaluated by a ROC curve. Experimental results indicate that the percentage of correct classification was 99.72%, obtaining 100% true positive (sensitivity) and 99.67% false positive (specificity), with the best classifier proposed. In case of the best classifier, we obtained a performance evaluation of classification of Az = 0.9875.
Abstract: Ant Colony Optimization (ACO) is a group of algorithms inspired by the foraging behavior of ant colonies in nature. Like their biological counterparts, a colony of artificial ants is able to adapt to the changes in their environment, such as exhaustion of a food source and discovery of a new one. In this paper, one of the basic ACO algorithms, the Ant System algorithm, was applied for edge detection where the edge pixels represent food for the ants. A set of grayscale images obtained by a nonlinear contrast enhancement technique called Multiscale Adaptive Gain is used to create a variable environment. As the images change, the ant colony adapts to those changes leaving pheromone trails where the new edges appear while the pheromone trails that are not reinforced evaporate over time. Although the images were used to create an environmental setup in which the ants move, the colonyâs adaptive behavior could be demonstrated on any type of digital habitat.
Abstract: This paper describes the use of networked control algorithms in designing a robotic swarm. The main goal of a robotic swarm is to divide one task into multiple simpler tasks. Have we designed a swarm this way, the main challenge would be the problem of delay in communication between individual robots. This paper also goes through the Swarm Intelligence concept and proposes the Network Formation Control algorithms to control a group of robots.
Abstract: Unmanned Aerial Vehicle (UAV) is defined as aircraft without the onboard presence of pilots. UAVs have been used to perform intelligence, surveillance, and reconnaissance missions. The UAVs are not limited to military operations, they can also be used in commercial applications such as telecommunications, ground traffic control, search and rescue operations, crop monitoring, etc. In this paper, we propose a swarm intelligence-based method for UAVs' route optimization. The team of UAVs is used for area coverage with the defined set of waypoints. The problem can be interpreted as a well-known Traveling Salesman Problem where the task is to find the route of minimal length such that all the waypoints are visited only once. We applied the Ant System algorithm and compared it with the Nearest Neighbor Search. The experimental results confirm the effectiveness of our method, especially for a large number of waypoints.
Abstract: Swarm Robotics refers to the application of Swarm Intelligence techniques where a desired collective behavior emerges from the local interactions of robots with one another and with their environment. In this paper, a modified Bees Algorithm is proposed for multi-target search and coverage by an autonomous swarm of robotic bees. The objective is to find targets in an unknown area, send their estimated locations and fitness values to other robots in swarm which then provide the coverage of the found targets in a self-organized, decentralized way. The robots are equipped with ultrasonic sensors for obstacle avoidance, thermal sensors for target detection, and ZigBee modules for local communication. For the experiments, a small swarm of robots was built to test the performance of the modified Bees Algorithm. The experimental results show that the swarm is self-organized, decentralized and adaptive, and it can be successfully applied to the unknown area search and coverage.
Abstract: New generations of Ground based Air Defense Systems of Systems used in modernized Armed Forces maintain the architectures used in their previous versions, constrained by hierarchical radio communications and centralized Command and Control restrictions. Current state-of-the-art technology in these areas allows a refocusing of these architectures towards a swarm approach. Advantages in terms of effectiveness, scalability and survivability are identified, and a preliminary set of swarm algorithms are proposed and analyzed. Upgrades of these preliminary algorithms are proposed as future research areas.
Abstract: Breast cancer is one of the leading causes to women mortality in the world. Cluster of Microcalcifications (MCCs) in mammograms can be an important early sign of breast cancer, the detection is important to prevent and treat the disease. In this paper, we present a novel method for the detection of MCCs in mammograms which consists of image enhancement by histogram adaptive equalization technique, MCCs edge detection by coordinate logic filters (CLF), generation, clustering and labelling of suboptimal features vectors by self organizing map (SOM) neural network. The experiment results show that the proposed method can locate MCCs in an efficient way.
Abstract: Conventional image edge detectors always result in missing parts of the edges. Broken edge linking is an image improvement technique that is complementary to edge detection, where the broken edges are connected to form closed contours in order to separate the regions in the image. In this paper, Ant System (AS) algorithm is modified for edge linking problem. As input, a binary image obtained after applying the Sobel edge operator is used. The proposed method defines a novel fitness function dependent on two variables: the grayscale visibility of the pixels and the length of the connecting edge, in order to obtain effective solution evaluation. Another novelty is of applying the grayscale visibility matrix as the initial pheromone trails matrix so that the pixels belonging to true edges have a higher probability of being chosen by ants on their initial routes, which reduces computational load. The results of the experiments are presented to confirm the effectiveness of the proposed method.
Abstract: This paper tests a novel improvement in neural network training by implementing Metaplasticity Multilayer Perceptron (MMLP) Neural Networks (NNs), that are based on the biological property of metaplasticity. Artificial Metaplasticity bases its efficiency in giving more relevance to the less frequent patterns and subtracting relevance to the more frequent ones. The statistical distribution of training patterns is used to quantify how frequent a pattern is. We model this interpretation in the NNs training phase. Wisconsin breast cancer database (WBCD) was used to train and test MMLP. Our results were compared to recent research results on the same database, proving to be superior or at least an interesting alternative.
Abstract: This paper presents new relevant results on the application of the optimization of backpropagation algorithm by a weighting operation on an artificial neural network weights actualization during the learning phase. This modified backpropagation technique has been recently proposed by the author, and it is applied to a multilayer perceptron artificial neural network training in order to drastically improve the efficiency of the given training patterns. The purpose is to modify the mean square error (MSE) objective function in order to improve the training efficiency. We show how the application of the weighting function drastically accelerates training convergence whereas it maintains neural network's (NN) performance.
Abstract: Many Artificial Neural Networks design algorithms or learning methods imply the minimization of an error objective function. During learning, weight values are updated following a strategy that tends to minimize the final mean error in the Network performance. Weight values are classically seen as a representation of the synaptic weights in biological neurons and their ability to change its value could be interpreted as artificial plasticity inspired by this biological property of neurons. In such a way, metaplasticity is interpreted in this paper as the ability to change the efficiency of artificial plasticity giving more relevance to weight updating of less frequent activations and resting relevance to frequent ones. Modeling this interpretation in the training phase, the hypothesis of an improved training is tested in the Multilayer Perceptron with Backpropagation case. The results show a much more efficient training maintaining the Artificial Neural Network performance.
Abstract: This paper presents some relevant results of a novel variant of the Backpropagation Algorithm to be applied during the Multilayer Perceptrons learning phase. The novelty consists in a weighting operation when the MLP learns the weights. The purpose is to modify the Mean Square Error objective giving more relevance to less frequent training patterns and resting relevance to the frequent ones. The inherent statistical distribution of training patterns is used to quantify how frequent a pattern is. The results, applied to a radar detector, show that Backpropagation with Weighting training requires much less training patterns maintaining the Artificial Neural Network performance.
Abstract: This work gives an overview of the broad field of computational swarm intelligence and its applications in swarm robotics. Computational swarm intelligence is modelled on the social behavior of animals and its principle application is as an optimization technique. Swarm robotics is a relatively new and rapidly developing field which draws inspiration from swarm intelligence. It is an interesting alternative to classical approaches to robotics because of some properties of problem solving present in social insects, which is flexible, robust, decentralized and self-organized. This work highlights the possibilities for further research.
Abstract: Animal swarms in nature are able to adapt to dynamic changes in their envi-ronment, and through cooperation they can solve problems that are crucial for their survival. Only by means of local interactions with other members of the swarm and with the environment, they can achieve a common goal more efficiently than it would be done by a single individual. This problem-solving behavior that results from the multiplicity of such interactions is referred to as Swarm Intelligence. The mathematical models of swarming behavior in nature were initially proposed to solve optimization problems. Nevertheless, this decentralized approach can be a valuable tool for a variety of applications, where emerging global patterns represent a solution to the task at hand. Methods for the solution of difficult computational problems based on Swarm Intelligence have been experimentally demonstrated and reported in the literature. However, a general framework that would facilitate their design does not exist yet. In this dissertation, a new general design methodology for Swarm Intelligence tools is proposed. By defining a discrete space in which the members of the swarm can move, and by modifying the rules of local interactions and setting the adequate objective function for solutions evaluation, the proposed methodology is tested in various domains. The dissertation presents a set of case studies, and focuses on two general approaches. One approach is to apply Swarm Intelligence as a tool for optimization and feature extraction, and the other approach is to model multi-agent systems such that they resemble swarms of animals in nature providing them with the ability to autonomously perform a task at hand. Artificial swarms are designed to be autonomous, scalable, robust, and adaptive to the changes in their environment. In this work, the methods that exploit one or more of these features are presented. First, the proposed methodology is validated in a real-world scenario seen as a combinatorial optimization problem. Then a set of novel tools for feature extraction, more precisely the adaptive edge detection and the broken-edge linking in digital images is proposed. A novel data clustering algorithm is also proposed and applied to image segmentation. Finally, a scalable algorithm based on the proposed methodology is developed for distributed task allocation in multi-agent systems, and applied to a swarm of robots. The newly proposed general methodology provides a guideline for future developers of the Swarm Intelligence tools.
DOI 
