Michal Puheim

Forward control of robotic arm using the information from stereo-vision tracking system

In this paper we present the feed-forward neural network controller of robotic arm, which makes use of tracking method applied to stereo-vision cameras mounted on the head of the humanoid robot Nao, in order to touch the tracked object. The Tracking-Learning-Detection (TLD) method, which we use to detect and track the object, is known for its state-of-art performance and high robustness. This method was adjusted to be usable with a stereo-vision camera system, in order to provide 3D spatial coordinates of the object. These coordinates are used as the input for the feed-forward controller, which controls the arm of a humanoid robot. The goal of the controller is to move the hand of the robot to the object by setting arm joints into position corresponding to the object location. The controller is implemented as an artificial neural network and trained using the error back-propagation algorithm. The experiment, which demonstrates the proof of the concept, is also denoted in this paper.

Interactive evolutionary optimization of fuzzy cognitive maps

Abstract Modeling dynamic systems with Fuzzy Cognitive Maps (FCMs) is characterized by the simplicity of the model representation and its execution. Furthermore, FCMs can easily incorporate human knowledge from the given domain. Despite the many advantages of FCMs, there are some drawbacks, too. The quality of knowledge obtained from the domain experts, and any differences and uncertainties in their opinions, has to be improved by different methods. We propose a new approach for handling incompleteness and natural uncertainty in expert evaluation of the connection matrix of a particular FCM. It is based on partial expert estimations and evolutionary algorithms in the role of an expert-driven optimization and outside of the FCM optimization (adaptation) research area known as Interactive Evolutionary Computing (IEC). In the present paper, a modification of IEC for the purposes of FCM optimization is presented, referred to as the IEO-FCM method, i.e., the Interactive Evolutionary Optimization of Fuzzy Cognitive Maps. Experimental results on two control problems suggest that the IEO-FCM method can improve the quality of an FCM even in situations without any measured data necessary for other known learning algorithms.

Three-term relation neuro-fuzzy cognitive maps

In this paper, we propose a novel approach to modeling using fuzzy cognitive maps, which we refer to as the Three-Term Relation Neuro-Fuzzy Cognitive Map or simply the TTR NFCM. The proposed method is mostly suited to model complex nonlinear technical systems with dynamic internal characteristics. With this method we aim to solve some of the most critical problems of the conventional fuzzy cognitive maps. We target two of these problems by hybridization with artificial neural networks. First of them is a linear nature of relations between the concepts. The second is a lack of mutual dependence between the relations connecting to the same concept. Finally, we tackle a problem of relation dynamics using an inspiration from the control engineering. While focusing on bringing these advanced additional methods to the design of cognitive maps, we also aim to keep the degree of dependency on expert knowledge on the same level as with the conventional fuzzy cognitive maps. We achieve this by utilizing the machine learning methods. However, since the proposed method is heavily dependent on automated data-driven learning, it is suitable mainly for systems which are well observable and can produce sufficient training datasets.

Normalization of inputs and outputs of neural network based robotic arm controller in role of inverse kinematic model

Goal of this paper is to discuss the methods usable to normalize inputs and outputs of the neural network controller used to control the arm of the humanoid robot with 3 degrees of freedom. The task of the controller is to solve the inverse kinematic problem, i.e. to move the hand of the humanoid robot to the target location given in arbitrary coordinate system other than its own kinematic chain defined by joint angle vector. In order to train accurate model for the controller it is necessary to normalize the values of input and output data in the training dataset. Data normalization within certain criteria, prior to the training process, is crucial to obtain satisfactory results as well as to fasten the training process itself. To proceed with the normalization we need to reduce domains of the training data in advance. Despite this task may look trivial, especially if I/O domains are clearly given, in some applications, such as finding the solution to the inverse kinematics problem of the humanoid robotic arm, it may become more complex and challenging. In this paper we will analyze possible options to perform normalization using expert oriented, automatic and hybrid approaches.

On practical constraints of approximation using neural networks on current digital computers

Goal of this paper is to highlight the most common problems and constraints which accompany the implementation of artificial neural networks on current digital computers. We focus on feed-forward multilayer neural networks, i.e. multilayer perceptrons, in role of function approximators. Multiple constraints of approximation by neural networks are discussed within the paper, taking into account research from the previous two decades. We address the issues of structural construction of feed-forward neural networks, learning and data pretreatment. Conclusions stated by universal approximation theorem cannot be blindly applied to implementations on real hardware without considering the limitations such as finite accuracy of floating point operations and data type overflow issues. This fact is emphasized in the paper.

Agent-Based Cloud Computing Systems for Traffic Management

The need for a safe and economically efficient traffic represents a challenge for creating traffic management systems. This paper deals with merging three basic concepts, namely cloud-based technologies, agent-based approaches and fuzzy cognitive maps to solve the localisation and path planing for several vehicles of different types in the form of independent agents, which communicate with a supervisory system located on the cloud. The proposed system was directly tested on a playground and obtained results were analysed using several selected criteria. Finally, some further possibilities of potential utilization and future research are mentioned.

Application of Tracking-Learning-Detection for Object Tracking in Stereoscopic Images

We use Tracking-Learning-Detection algorithm (TLD) [1]-[3] to localize and track objects in images sensed simultaneously by two parallel cameras in order to determine 3D coordinates of the tracked object. TLD method was chosen for its state-of-art performance and high robustness. TLD stores the object to be tracked as a set of 2D grayscale images that is incrementally built. We have implemented the 3D tracking system into a PC, communicating with the Nao humanoid robot [4][5] equipped with a stereo camera head. Experiments evaluating the accuracy of the 3D tracking system are presented. The robot uses feed-forward control to touch the tracked object. The controller is an artificial neural network trained using the error Back-Propagation algorithm. Experiments evaluating the success rate of the robot touching the object are presented.

Intelligent space at center for intelligent technologies — system proposal

In this paper we present an engineering proposal for a data processing system based at the Center for Intelligent Technologies. The proposed system forms the Intelligent Space using the collection of sensors including IP cameras, Kinect sensors and other. The proposed data processing infrastructure is based on the Fog Computing paradigm established by Cisco. The main reason for utilization of this approach is the necessity to process large amounts of data produced by the sensors within the considered Intelligent Space. Such amount could not be processed using typical Internet of Things solutions which rely heavily on remote Cloud Computing. The paper provides detailed description of hardware, software and networking solutions and also provides a couple of example applications within the implemented Intelligent Space.

Efficient FCM computations using sparse matrix-vector multiplication

Fuzzy cognitive maps (FCM) are often represented and implemented using matrix-vector multiplication (MxV). Since the multiplication operation is critical to the performance of the FCM computations, it is important to secure its efficient implementation. Considering the connection matrix used to represent the FCM is often static and since it often contains only several nonzero elements, it is viable to transform it into another particular representation suitable to perform sparse matrix-vector multiplication (SpMxV). This paper shows a performance benchmark for the most common SpMxV representations, namely the CRS and CCS. It also examines the sparsity threshold at which it is more efficient to use naïve dense MxV.

Automatic predictor generator & behaviour rule extractor — A system proposal

In this paper we present a proposal for a data-mining system deployed as a cloud service which is supposed to be used for a big data analysis. The main purpose of the system is the analysis of a vast number of event logs using means of data aggregation, clustering, classification and prediction. The system is composed of two components implemented as software services. The Automatic Predictor Generator is supposed to provide a meaningful way to aggregate large amounts of data and the Automatic Behavior Rule Extractor deals with proper analysis of these aggregations. Results of the system are the prediction rules usable for support of decision-making and in areas such as management, marketing, customer segmentation, classification, behavior prediction etc.