We report on a trans-Atlantic PhD student mobility program that connects two graduate research training initiatives in the US and Ireland, centered on developing future researchers in artificial intelligence (AI) and machine learning (ML). We discuss both the structure of the student exchange experiences and share key learnings from this international collaboration. The most important lesson learned is that providing a structured mobility program and matched visiting pairs is a highly effective way to improve learning outcomes compared to more typical ad-hoc individual visits.

Effective human–robot collaboration (HRC) requires robots to understand and adapt to humans' psychological states. This research presents a novel approach to quantitatively measure human comfort levels during HRC through the development of two metrics: a comfortability index (CI) and an uncomfortability index (UnCI). We conducted HRC experiments where participants performed assembly tasks while the robot's behavior was systematically varied. Participants' subjective responses (includingsurprise,anxiety,boredom,calmness, andcomfortabilityratings) were collected alongside physiological signals, including electrocardiogram, galvanic skin response, and pupillometry data. We propose two novel approaches for estimating CI/UnCI: an adaptation of the emotion circumplex model that maps comfort levels to the arousal–valence space, and a kernel density estimation model trained on physiological data. Time-domain features were extracted from the physiological signals and used to train machine learning models for real-time comfort levels estimation. Our results demonstrate that the proposed approaches can effectively estimate human comfort levels from physiological signals alone, with the circumplex model showing particular promise in detecting high discomfort states. This work enables real-time measurement of human comfort during HRC, providing a foundation for developing more adaptive and human-aware collaborative robots.

Our AI-focused traineeships for graduate students integrate research and education components to contribute to diversifying the AI research workforce. We describe the program and introduce multiple strategies to achieve interdisciplinarity, diversity, equity, inclusion, and accessibility. Early evaluation results are included.

This work evaluates a Continuous Wave (CW) Time-of-Flight (ToF) camera, Stereoscopic camera, and LiDAR to determine if they are potential candidates for point-rich on-robot sensing in Speed and separation monitoring (SSM) applications. These experiments characterize the static and dynamic behaviors of the sensors while mounted on-robot. From these tests, it was found that ToF and Stereo cameras exhibit better performance to their more expensive LiDAR counterpart. Specifically, it was observed that the ToF camera demonstrated better depth accuracy while the Stereo camera generated better 3D reconstruction accuracy. Overall, ToF and Stereo Cameras demonstrate that with continued innovation and integration, these sensors could become the building blocks to point rich on-robot SSM.

We introduce three new multi-modal data sets. They contain physiological and/or emotional information about human interactions with robotic arms in proximity to completing a task in an industrial setting. The data sets provide data from human subjects engaged in the assistive task of assembling a PVC joint pipe with robots. These data streams were collected to analyze and improve the comfort and safety of humans collaborating with robots in proximity in an industrial setting. These data sets can appeal to researchers studying human-robot collaboration, robot adaptation, and affective computing. Our data is stored in various formats, including images and human-readable Comma-Separated Values (CSV) or JavaScript Object Notation (JSON) files.

Many attempts have been made at estimating discrete emotions (calmness, anxiety, boredom, surprise, anger) and continuous emotional measures commonly used in psychology, namely ‘valence’ (The pleasantness of the emotion being displayed) and ‘arousal’ (The intensity of the emotion being displayed). Existing methods to estimate arousal and valence rely on learning from data sets, where an expert annotator labels every image frame. Access to an expert annotator is not always possible, and the annotation can also be tedious. Hence it is more practical to obtain self-reported arousal and valence values directly from the human in a real-time Human-Robot collaborative setting. Hence this paper provides an emotion data set (HRI-AVC) obtained while conducting a human-robot interaction (HRI) task. The self-reported pair of labels in this data set is associated with a set of image frames. This paper also proposes a spatial and temporal attention-based network to estimate arousal and valence from this set of image frames. The results show that an attention-based network can estimate valence and arousal on the HRI-AVC data set even when Arousal and Valence values are unavailable per frame.

In this work, we focus on multi-step manipulation tasks that involve long-horizon planning and considers progress reversal. Such tasks interlace high-level reasoning that consists of the expected states that can be attained to achieve an overall task and low-level reasoning that decides what actions will yield these states. We propose a sample efficient Previous Action Conditioned Robotic Manipulation Network (PAC-RoManNet) to learn the action-value functions and predict manipulation action candidates from visual observation of the scene and action-value predictions of the previous action. We define a Task Progress based Gaussian (TPG) reward function that computes the reward based on actions that lead to successful motion primitives and progress towards the overall task goal. To balance the ratio of exploration/exploitation, we introduce a Loss Adjusted Exploration (LAE) policy that determines actions from the action candidates according to the Boltzmann distribution of loss estimates. We demonstrate the effectiveness of our approach by training PAC-RoManNet to learn several challenging multi-step robotic manipulation tasks in both simulation and real-world. Experimental results show that our method outperforms the existing methods and achieves state-of-the-art performance in terms of success rate and action efficiency. The ablation studies show that TPG and LAE are especially beneficial for tasks like multiple block stacking. Additional experiments on Ravens-10 benchmark tasks suggest good generalizability of the proposed PAC-RoManNet.

Industrial robots working in isolation in a highly automated system are valued for their high productivity. The shortcomings of these pure robotic cells become more apparent when flexibility in production is required to respond to varying production volumes and customized product demands. Complete automation is highly productive, but it is costly to set up and difficult to change. On the other hand, manual production, although flexible, is slower and prone to human errors. Hence, in industry, smarter automation methods that leverage the dexterity, flexibility, and decision-making capability of a human to speed, precision, and power of a robot are required. In industry, the need for flexibility in production has resulted in the acceptance of human-robot collaboration (HRC) as a viable alternative. The objective of this survey is to address the main challenges in HRC (safety, trust-in-automation, and productivity), safety measures, types of HRC, technical standards, and conceptual categorization of HRC: awareness, intelligence, and compliance.

In this paper, we present a modular robotic system to tackle the problem of generating and performing antipodal robotic grasps for unknown objects from the n-channel image of the scene. We propose a novel Generative Residual Convolutional Neural Network (GR-ConvNet) model that can generate robust antipodal grasps from n-channel input at real-time speeds (~20ms). We evaluate the proposed model architecture on standard datasets and a diverse set of household objects. We achieved state-of-the-art accuracy of 97.7% and 94.6% on Cornell and Jacquard grasping datasets, respectively. We also demonstrate a grasp success rate of 95.4% and 93% on household and adversarial objects, respectively, using a 7 DoF robotic arm.

In this paper, a new design for a soft robotic gripper is proposed which combined humanoid fingers with bioinspired suction cups. The gripper is capable of reconfiguration of the fingers to allow multiple grasping strategies to be achieved. The grasping strategies implemented are cylindrical, spherical, and tip gripping of an object. Construction of the physical soft robotic gripper was implemented in hardware and software to allow for experimentation of the new design. Experiments of grasping objects of varied shape and weight were done ans results tabulated. The experiments gave a deeper understanding of the design choices impact on the ability of the gripper to perform pick and place tasks.

In this paper, an analysis of the sensing volume coverage of robot workspace as well as the shared humanrobot collaborative workspace for various configurations of onrobot Time-of-Flight (ToF) sensor array rings is presented. A methodology for volumetry using octrees to quantity the detection/sensing volume of the sensors is proposed. The change in sensing volume coverage by increasing the number of sensors per ToF sensor array ring and also increasing the number of rings mounted on robot link is also studied. Considerations of maximum ideal volume around the robot workspace that a given ToF sensor array ring placement and orientation setup should cover for safe human robot interaction are presented. The sensing volume coverage measurements in this maximum ideal volume are tabulated and observations on various ToF configurations and their coverage for close and far zones of the robot are determined.

In this paper, a framework for monitoring human physiological response during Human-Robot Collaborative (HRC) task is presented. The framework highlights the importance of generation of event markers related to both human and robot, and also synchronization of data collected. This framework enables continuous data collection during an HRC task when changing robot movements as a form of stimuli to invoke a human physiological response. It also presents two case studies based on this framework and a data visualization tool for representation and easy analysis of the collected data during an HRC experiment.

In this paper we apply generative modeling to Gaussian Mixture Models (GMM) as a solution to high dimensional novel event detection and analysis on radio frequency (RF) power generators. A family of Denoising and β-Disentangling Variational Autoencoders (β-DVAE) is used to encode datasets into lower and more salient dimensions. The reduced feature sets are modeled using GMMs where model parameters are learned using the Expectation Maximization (EM) algorithm. The data is obtained from two different generator models operating under normal as well as known not-normal conditions. This approach is also tested on standard classification sets. Robust testing is reported to achieve a target class accuracy of 98.16% for the target RF generator. Additionally, the GMM parameters are decoded by the generative model to the original data space for calculating per-variable fitness in order to provide an initial novel event analysis for engineers and technicians. This per-variable fitness is very critical for determining each variable's contribution to a novelty so that engineers can perform informed trouble shooting and maintenance of the RF generators.

In this paper, a range sensing setup for performing detection and monitoring in an industrial robot workspace is presented. The setup uses a ring of Time-of-Flight laser range sensors mounted on a robot. A 3D simulation setup with the properties of the sensor mounted on a UR10 robot and a simple pick and place task with a human avatar is used to analyze its behavior and viability with industrial arm robots. Collision detection strategies based on human-robot separation distance and relative speeds are also implemented. These strategies are evaluated based on the human safety, robot performance and productivity of the task. The parameters and results of the experiments are tabulated. The results show the benefits of achieving dynamic awareness of the robot in comparison with the conventional methods used in industry. The development of the prototype sensor ring is also shown and the future work discussed.

In this paper, a new low computational complexity machine learning classifier is developed to detect eye movements from an Electrooculography (EOG) signal. The proposed classier is based on finding the maximum relative global peak of a set of signals that are decomposed from the original EOG signals. Unlike many other classifiers, the proposed classifier does not depend on threshold signal levels or spectrum analysis of the EOG signals which require nonlinear and complex operations. The proposed classifier uses four predefined functions which are divided into two groups to distinguish between the horizontal and vertical eye movements. The new proposed classifier can be used for on low powerful boards because it does not require high speed processors where it needs to calculate only one optimum parameter that maximizes the global accuracy. The performance of the proposed classifier is better and faster by 40% than the SVM classifier. In addition, there is a misclassification between two classes only rather than three classes.

In this paper, a system to aid the visually impaired by providing contextual information of the surroundings using 360° view camera combined with deep learning is proposed. The system uses a 360° view camera with a mobile device to capture surrounding scene information and provide contextual information to the user in the form of audio. The scene information from the spherical camera feed is classified by identifying objects that contain contextual information of the scene. That is achieved using convolutional neural networks (CNN) for classification by leveraging CNN transfer learning properties using the pre-trained VGG-19 network. There are two challenges related to this paper, a classification and a segmentation challenge. As an initial prototype, we have experimented with general classes such restaurants, coffee shops and street signs. We have achieved a 92.8% classification accuracy in this paper.

The field of Sign Language Recognition (SLR) has become an increasingly popular research topic. The goal of this study is an SLR system that will be capable of identifying a subset of 50 of the most common American Sign Language (ASL) word signs using surface electromyography and accelerometer data for multiple signers. All data was collected from deaf, fluent ASL users. A windowing approach is used with different time domain features for feature extraction. The samples are divided into one and two-handed signs, each of which are used to train a Support Vector Machine classifier. Samples from all but one subject are used to train the classifiers. The classifiers are then tested on both data held out from the subjects used for training and the subject that was left out. The resulting system had an average accuracy of 59.96% for trained subjects and 33.66% for the subject left out. To compare this approach to others, 40-word and 7-word sign sets are trained and tested using this method. The proposed system performed comparably with literature for the 40-word set, and better for the 7-word set.

A novel hexapod-quadcopter robot was developed with multicopter flight hardware directly embedded in the hexapod legs. This paper discusses the locomotion implementations for walking and flying in addition to transitional procedures necessary to switch between these two modes of operation. The algorithms were programmed using Robot Operating System on the 3D printed robot and tested to verify feasibility and performance.

Human detection systems are becoming more important as more automatic and robotic systems are being used in the world. RGB, depth, and thermal images can be used together to produce a better detection system that works in situations where one of the sensors might not produce valuable data. HOG features can provide valuable information for detecting humans in an image and that data can be used to train individual classifiers to detect humans in a scene. The combination of sensor modalities in conjunction with individual classifiers can create a human detection system that can detect partially obscured humans. The multi-layer classifier that was created provided a high level of accuracy when tested against untrained data. The multi-layer classifier performed better than eleven of the twelve individual classifiers, but did not overcome the SVM thermal HOG classifier. The multi-layer classifier had a much tighter standard deviation and fell within the band of the SVM thermal classifier.

A Human Action recognition system is proposed using the Skeletal Tracking from Kinect. The angular information of the joints helps in handling scaling errors. Vectors are generated using the joint coordinates and the angles of each joint are used as features for key pose recognition. A rotational compensation is included in the feature to handle rotational errors. The key poses are recognized using Similarity Matching Technique, neural network and decision tree algorithms. The recognized key postures are fed into a decision forest to pick the action based on the trained sequence of key poses.

Sign Language Recognition (SLR) system is a novel method that allows hard of hearing people to communicate with society. In this study, an American Sign Language (ASL) recognition system was proposed by using the surface Electromyography (sEMG). The objective of this study is to recognize the American Sign Language alphabet letters and allow users to spell words and sentences. For this purpose, sEMG signals are acquired from subject's right forearm for 27 American Sign Language gestures, 26 English alphabet letters, and one for home position. Time domain, frequency domain (band power), power spectral density (band power), and average power features were used as the feature extraction methods. After feature extraction, Principal Component Analysis (PCA) was applied to obtain uncorrelated features. As a classification method, Support Vector Machine and Ensemble Learning algorithm were used and their performances were compared with tabulated results. In conclusion, the results of this study show that sEMG signal can be used for SLR systems.

Sign Language Recognition (SLR) system is a method which allow deaf people to communicate with society. In this study, Real-Time Sign Language recognition system was proposed by using the surface Electromyography (sEMG). To this purpose, sEMG data acquired from subject right forearm for all twenty six American Sign Language gestures. Raw sEMG data was filtered, feature extracted and fed into classification. Support Vector Machine (SVM) with one vs. all approach was used for multi class classification. The experiment result of offline system is reaching a recognition rate of 91.% accuracy and real-time system has a recognition rate of 82.3% accuracy. The results of the proposed system shows that sEMG signal can be used for Real-Time SLR systems.

In this paper, we introduce a framework for a system which intelligently assigns an edge detection filter to an image based only on features taken from the image. The system has four parts, the training set which consists of an image and its edge image ground truth, feature extraction, training filter creation, and system training. A prototype system of this framework is given. In the system feature extraction is performed using a GIST methodology which extracts color, intensity, and orientation information as features. The set of image features are used as the input to a single hidden layer feed forward neural network trained using back propagation. The system trains against a set of linear Cellular Automata filters which are determined to best solve the edge image according to the Bad delay Delta Metric. This metric takes into account false positives and false negative error by scaling the errors relative to the perpendicular distance that they are off from the ground truth. The system was trained and tested against the images from the Berkeley Segmentation Database. The results from the testing indicate that the system performs better than standard methods in many cases but on the whole is only on par across a wide range of images.

In this paper we apply a specific machine learning technique for classification of normal and not-normal operation of RF (Radio Frequency) power generators. Pre-processing techniques using FFT and bandpower convert time-series system signatures into single feature vectors. These feature vectors are modeled using k-component Mixture of Gaussians (MoG) where components and corresponding parameters are learned using the Expectation Maximization (EM) algorithm. Data is obtained from three different generator models operating under normal and multiple different not-normal conditions. Exploration into algorithmic parameter effects is conducted and empirical evidence used to select sub-optimum parameters. Robust testing is reported to achieve a 3s classification accuracy of 95.91% for the targeted RF generator. Additionally, a custom C++ library is implemented to utilize the learned model for accurate classification of time-series data within an embedded environment such as a RF generator. The embedded implementation is reported to have a small storage footprint, reasonable memory consumption and overall fast execution time.

The twin rotor MIMO system (TRMS) is a helicopter-like system that is restricted to two degrees of freedom, pitch and yaw. It is a complicated nonlinear, coupled, MIMO system used for the verification of control methods and observers. This paper compares the ability of three adaptive sliding mode controllers (SMC) to suppress chattering. Once the design is complete the controllers are implemented in simulation and experimentally. The performance of the controllers are compared against a PID controller and other controllers in the literature. The ability of the sliding mode controllers (SMC) to suppress chattering is also be explored.

In this paper we apply various machine learning techniques for fault detection of RF (Radio Frequency) Power Generators. Fast Fourier Transform features are used in our analysis for all experiments. Radial Basis Function Networks (RBF) is used to build a two class classifier to differentiate between normal and one fault condition. We apply three one class classifiers to model the normal operating conditions. The data is obtained from five different generators of the same model type.

Block matching algorithms successfully reduce the computational load of the motion estimation. Block based motion estimation algorithms rely on block matching of the macro blocks defined in the search windows. In this paper, particle swarm optimization (PSO) is applied on block-matching problem. Particle swarm optimization is a computationally efficient iterative search algorithm that finds the best matching block between two consecutive frames. Finding the moving block in the image sequence requires a minimization algorithm. Using particle swarm optimization, the computational cost of the block matching is aimed to decrease. Particle swarm optimization is a heuristic algorithm and necessarily dependent on the selections of the parameters assigned. In our case, the particle swarm algorithm is designed to work on a small subset of the population of blocks. The position of a macro block in the estimated image can be revealed by the motion vectors generated by the particle swarm optimized block.

In this paper we discuss the application of two-dimensional linear cellular automata rules to the problems of edge detection in monochromatic images. We proposed an efficient and simple method of edge detection based on uniform cellular automata transition matrix representation. We investigate of cellular automata linear rules for edge detection by using matrix representation, in some cases they are strong and some other rules are not in fact useful for practical edge detection. All rules give computationally effective result since the computation mechanism consist only matrix multiplication. Finally, we present some results of the proposed linear rules for edge detection and compare with some classical results.

In this paper we apply two variable elimination algorithms to data obtained from an RF (Radio Frequency) Power Generator Fault Mode for analysis. We use a two wrapper approach using Support Vector Machines (SVM) and Radial Basis Function Networks (RBF) to build an efficient classifier with variable elimination. Comparisons are made for both continuous and discrete datasets.

EMG based input device is a natural means of human computer interface (HCI) because the electrical activity induced by the human's arm muscle movements can be interpreted and transformed into computer's control commands. In this paper, we describe an approach for classifying electromyography (EMG) signals using a multilayer perceptron neural network (MLP) and Bayesian classifier (BC) with the wavelet transformation technique for feature selection to discriminate 6 classes of motions to control a mouse. Wavelet Transformation (WT) was applied to raw EMG data and in order to decrease the dimension of the feature sets, principle component analysis (PCA) and sequential forward selection (SFS) were utilized.

This paper explores different crossover operators and how they affect the Scatter Search (SS) algorithm in unconstrained function optimization. It also explores how the size of the reference set affects convergence and robustness. An introduction to Scatter Search is given along with the typical template. It follows with a thorough explanation of the specific implementation that is used. Two types of tests are designed: (1) what is the best fitness value that it can reach within a maximum number of fitness evaluations and (2) how fast it can converge to goal fitness. These tests are performed using different crossover operators and reference set sizes. Results are also compared apples to apples with a basic implementation of Particle Swarm Optimization (PSO). The results show that although PSO is quicker at converging, SS shows more robustness with higher overall success rates.

The road network of a small town is represented by a directed graph. Road junctions are the vertices of this graph and each road segment (which has a length and a priority value) is represented by a directed edge. Priority values are numbers 1, 2, etc. with the assumption that 1 is the highest priority. We seek to compute an optimal route map that begins at a particular vertex (the depot) and covers all the edges at least once and returns to the start vertex. The parameters that we wish to minimize are: the total distance covered (thereby minimizing the deadhead miles), the number of u-turns and priority misplacements. In this paper, we propose a Genetic Algorithms-based solution to compute near-optimal route maps in such a graph. Specifically, we have developed a Java software application that generates route maps that minimize a linear combination of the three parameters. We have experimented with reasonably large graphs and obtained good solutions. These solutions are especially useful in snowplow routing for small towns, as plowing costs consume significant portions of the total municipal budgets of these communities. Most of the route planning is currently done manually and routes have evolved over time by experience. In these times of severe budget stress, route planning using our approach can help in performing this essential service in an efficient manner.

This paper analyzes the application of different classification techniques for Electroencephalography (EEG) signals. Fuzzy Functions Support Vector Classifier (FFSVC), Improved Fuzzy Functions Support Vector Classifier (IFFSVC) and a novel hybrid technique that has been designed utilizing Particle Swarm Optimization and Radial Basis Function Networks (PSO-RBFN) have been studied. The classification performance of the techniques is compared on the same standard datasets that are publicly available and used by many Brain Computer Interface (BCI) researchers. Results show that proposed classifiers might reach the classification performance of state of the art classifiers and might be used as alternative techniques in the classification applications of EEG signals.

This paper extends an XML based system of systems simulation framework using DEVS to support hardware-in-the-loop simulations for SoS. A system of systems approach enables the simulation and analysis of multiple complex, independent, and cooperative systems by concentrating on the data transferred among systems. This paper wraps information exchanged among heterogeneous systems in XML to enable receiving systems to correctly parse and interpret the information. A Groundscout robot is deployed as a real agent working cooperatively with virtual agents in a robotic swarm. The DEVS activity concept facilitates communication between the real system and virtual systems in the SoS. A robust threat detection example is provided. Initial performance metrics of the SoS are briefly discussed.

This paper presents a system of systems approach to threat detection through integration of heterogeneous independently operable systems. The approach is presented on a realistic situation where a human-controlled base robot, swarm robot(s), and sensors work together to obtain a decision about a possible threat in the environment. The base robot is remotely operated by a human using a haptic control system. The swarm robot(s) are autonomous and can accept directives from the base robot. Finally, sensors directly communicate with (report to) the base robot. In this scenario, heterogeneous systems and human interact in a system of systems architecture. With the inclusion of human expert and sensor verification of swarm robots, the system can successfully perform the threat detection and reduce the false alarms. Finally, a system of systems simulation framework including a base robot, a swarm robot, and two sensors is presented in addition to an experimental evaluation of the proposed SoS architecture.

This research aims at characterizing and predicting the Young's modulus of thin film materials that are utilized in the microelectromechanical systems (MEMS). As a proof of concept, aluminum and TEOS thin films were analyzed using bilayer cantilever as a test structure. Due to the lack of understanding of the mechanical behavior of thin film materials in the micro-scale domain, empirical models were developed that utilize soft computing techniques. As a result, this methodology is foreseen to be an essential tool for MEMS designers as it can estimate and predict effective Young's modulus of materials in the micro-scale domain. In the estimation phase, 2D search and micro genetic algorithm were studied and in the prediction phase, back propagation based neural networks and one dimensional radial basis function networks (1D-RBFN) were studied. All combinations of these soft computing techniques are evaluated. Based on the results, we conclude that among the various combinations tested, the combination of 1D-RBFN (prediction phase) and GA (estimation phase) presented the best results. Research is in progress in applying other algorithms such as support vector machines as well as investigating other novel test structures that can be used to extract other material properties such as coefficient of thermal expansion.

This paper describes the design and development of a ping-pong robotic arm as an application of robotic vision. Displaced frame difference (DFD) is utilized to segment the ball motion from background motion. 3-D ball tracking using parametric calibration of single CCD camera is explained. This visual information is temporally updated and further employed to guide a robot arm to hit the ball at a specified location. The results signify the system development based on single camera tracking. System latency is measured as a function of the camera interface, processor architecture, and robot motion. Various hardware and software parameters that influence the real time system performance are also discussed.

This paper presents an efficient algorithm for learning a Bayesian belief network (BBN) structure from a database, as well as providing a comparison between two BBN structure fitness functions. A Bayesian belief network is a directed acyclic graph representing conditional expectations. In this paper, we propose a two-phase algorithm. The first phase uses asymptotically correct structure learning for efficient search space exploration, while the second phase uses greedy model selection for accurate search space exploration. The minimum description length (MDL) structure fitness function is also compared with the database given model probability (DGM) fitness function in the second phase. The model selection algorithms are applied to the ALARM network to provide a comparison for the accuracy of the techniques.

Navigation based on cognition is seen in many instances in the animal community. Maps are very useful for navigation in unknown and complex environments. User defined and preinstalled maps can be very useful in navigation in complex environments, while adaptively built maps become very essential in unknown environments. Intelligence is bestowed in the effective utilization of data to produce decisions. In the case of navigation, intelligence is seen when information collected en-route from a source to a destination is used judiciously in reaching the destination. In this paper we deal with the foraging of a piece of terrain as a process of navigation. The application here is on mine detection. Mines are placed over a field at unknown locations and swarm intelligence based agents are deployed for demining them. The efficiency of the demining process is measured in terms of the quickness with which they can clear the mines and the assurance that the minefield is clear of all of the mines at the end of the process. Simulation results acknowledging the performance under different working conditions of the swarm robots are presented. In an endeavor to substantiate our claims in simulation we decided to build swarm robots performing the task on an artificially developed environment.

A graph is edge-biconnected if it requires the removal of at least two edges to disconnect it. Assume that we have weighted graph that is not biconnected, and an additional set of augmentation edges. The (NP-hard) edge biconnectivity augmentation problem is to select a minimal subset of the augmentation edges, whose inclusion will cause the graph to be biconnected. This paper explores the application of particle swarm optimization and genetic algorithms for this problem.

The field of robotics is in rapid development. As robots become cheaper to build, new applications involving many robots systems can be envisioned. One reason for using many robots is to achieve robustness. Having many robots, however, does not ensure robustness. A control strategy and robot behaviors must be engineered to incorporate robustness into the system. Swarm intelligence based approaches are popular for developing optimal and robust control strategies for systems of robots. Here we analyze the behavior of a swarm of robots modeled after a swarm of ants, where tasks are spatially distributed in the environment and robots/ants are recruited through short-range recruitment. For ants that move probabilistically in response to the short range signal and who adjust their probabilities such that they are near a phase change boundary, or edge-of-chaos, in the mean field analysis of their motions, we find a significant improvement in the robustness of the system.

Multi-agent based solutions to problems, whether they are software agents or physical robots, are attractive because they are robust and scalable. Fundamental aspects of designing multi-agent systems involve modeling the intelligence of the agents and modeling their interactions. The intelligences of agents modeled here are encoded in Bayesian representations of their world. The agents interact only by observing others and moving in such a way so as to probabilistically maximize their internalized goal or utility. Using this multi-agent framework, a three agent herding is explored.

The paper proposes a structural Bayesian network learning in a biological decision-theoretic intelligent agent model to solve a herding problem. The proposed structural learning methods show that an agent can update its world model by changing the structure of its Bayesian network with the data gathered by experience. The structural learning of the Bayesian network is accomplished by implementing a score based greedy search algorithm. The search algorithm is designed heuristically and exhaustively. A complexity analysis for the search algorithms is performed. Intelligent agent software, IntelliAgent, is written to simulate the herding problem with one sheep and one dog.

An artificial immune system is applied to a pattern recognition problem. Artificial immune systems (AIS) possess nonlinear classification properties along with the biological properties such as self/non-self identification, positive and negative selection. In this paper, we propose a computational implementation of negative selection of immune system along with genetic algorithm to perform a color image classification task. The images used for classification are finished wooden components. The wooden components are kitchen cabinets manufactured by American Woodmark Corporation. The images are obtained from Virginia Polytechnic Institute and State University. The classification addresses the average RGB values of the images and the percentage of correct classification is obtained for a set of test and sample images.

Proposes a biological decision-theoretic intelligent agent model to solve a herding problem. The proposed intelligent agent model is designed by combining Bayesian networks and influence diagrams. In our agent design, we used Y. Shoham's (1993) agent-oriented programming paradigm that defines an intelligent agent by its belief, preference and capabilities. Intelligent agent software is written to realize the proposed intelligent agent model. The same software is then used to simulate the herding problem with one sheep and one dog. Simulation results show that the proposed intelligent agent is successful in establishing a goal (herding) and learning other agents' behaviors.