This manuscript analyses the performance of different machine learning models classifying hand gestures from electromyography (EMG) signals. The EMG information is obtained from the WyoFlex armband, a wearable bracelet with four sensors that capture the EMG of the forearm. This study considers four different models, the K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Long Short-Term Memories (LSTM) for classifying four and six hand gestures. The methodology comprises different scenarios, including the application of the Synthetic Minority Over-Sampling (SMOTE) algorithm to increase the cardinality of the dataset and the Minimum-Redundancy-Maximum-Relevance (MRMR) technique to reduce computational costs. These models are compared considering the classification of four and six different hand gestures in three different scenarios: a) including the SMOTE technique, b) including the MRMR procedure without SMOTE, and c) including both MRMR and SMOTE. The results show an overall accuracy between 79.32% to 94.90% on the four gestures classification and from 75.14% to 92.80% for six gestures classification, with the SVM classifier producing the best performance in training time and accuracy. The application of the MRMR algorithm yields a reduction of the training time in all models applying the K-fold and Leave-One-Subject-Out (LOSO) cross validation methods, the accuracy is not strongly affected reducing the number of features in almost all the models.

This work presents the design of a control trajec-tory algorithm based on Barrier Lyapunov functions (BLF) for a class of third-order electro-mechanical systems. The method considers unknown control gains and uncertain plant parame-ters, a common problem in control engineering applications. In order to deal with the uncertain parameters, this study proposes a backstepping adaptive strategy applying Nussbaum functions to know the sign of the control gains. The BLF imposes partial constraints to keep the trajectories of the system inside a safe operation set. The proposed controller is tested in a virtual one-link robotic manipulator including actuator dynamics to develop an adequate trajectory tracking. The maximum tracking error norm was 0.02, which fulfilled the maximum constrained bound of 0.5.

Over the last decade, the research in autonomous robots has increased the development of mechanisms, navigation, and control schemes. Robots constitute capabilities for doing specific human tasks depending on the mechanism and environment involved. For mobile robots, diverse techniques formulated navigation and control algorithms, some of them have applied sliding modes and other modern robust control techniques. Actually, most control algorithms based their development in the kinematic model. This aims to control a car-like robot mobile robot, with a backstepping strategy. At each step of the bacstepping procedure a second order super-twistng sliding mode algorithm force the states of the kinematic error, after a nonlinear transformation, to zero in finite-time. The proposed controller improves the tracking trajectory task. Numerical results demonstrate the effectiveness of the algorithm. A comparison with a classical proportional-integral-derivative controller enhances the advantages of applying a sliding mode strategy.

This manuscript proposes the design of a speech synthesis algorithm based on measured electroencephalographic (EEG) signals previously classified by a class of neural network with continuous dynamics. A novel multilayer differential neural network (MDNN) classifies a database with the EEG studies of 20 volunteers. The database contains information described by input-output pairs corresponding to EEG signals and a corresponding word imagined by the volunteer. The suggested MDNN estimates the unknown relationship between the information instances and suggests the most-likely word that the user wants mentioning. The proposed MDNN satisfactory classifies over 95% a set of words obtained from the suggested EEG study in which, the users have to watch four different geometric figures on a screen.

This manuscript presents the design of a deep differential neural network (DDNN) for pattern classification. First, we proposed a DDNN topology with three layers, whose learning laws are derived from a Lyapunov analysis, justifying local asymptotic convergence of the classification error and the weights of the DDNN. Then, an extension to include an arbitrary number of hidden layers in the DDNN is analyzed. The learning laws for this general form of the DDNN offer a contribution to the deep learning framework for signal classification with biological nature and dynamic structures. The DDNN is used to classify electroencephalographic signals from volunteers that perform an identification graphical test. The classification results show exponential growth in the signal classification accuracy from 82 percent with one layer to 100 percent with three hidden layers. Working with DDNN instead of static deep neural networks (SDNN) represents a set of advantages, such as processing time and training period reduction up to almost 100 times, and the increment of the classification accuracy while working with less hidden layers than working with SDNN, which are highly dependent on their topology and the number of neurons in each layer. The DDNN employed fewer neurons due to the induced feedback characteristic.

Unsupervised learning is a target free methodology to classify unorganized information. This study proposes a new unsupervised learning method for classifying unlabeled targets based on convex ellipsoidal sets. The method described here uses ellipsoidal calculus tools to realize pattern classification via a clustering scheme. The algorithm consisted in adjusting the number of clusters defined by an ellipsoidal set as well as their centers, shape forms and orientation. All these processes were realized without preliminary information on the data distribution. The application of an inner gradient descent algorithm permitted the adjustment of these parameters, using the standard deviation of data in the cluster with respect to the semi-axis of ellipsoid that contains the corresponding data. A specific evaluation of the proposed algorithm used two-dimensional attributes databases. The outcomes of the classification results were compared with those produced by the K-means and DBSCAN algorithm. Comparable classification results were achieved overcoming some drawbacks of the methods.

This manuscript presents an algorithm to classify continuously electroencephalographic (EEG) signals based on deep differential neural networks (DDNNs). The learning laws are obtained by the second stability method of Lyapunov that requires the solution of a set of matrix differential equations. The robustness of this technique allows the analysis and classification of bio-signals like EEG signals. The EEG signals have complex dynamics and they are strongly affected by noises in the measurements and a high degree of variability between different studies in patients. The main strength of DDNNs is their feedback property, which allows them to work with the time-dependent variation of the EEG signals. The DDNNs are tested in a database constituted of EEG signals acquired from a study made in ten volunteers. The study consisted of the acquisition of EEG measurements of the volunteers recognizing geometrical figures appearing in a graphic user interface. The DDNN obtained better performance than a single layer differential neural network and a convolutional neural network.

This manuscript deals with the problem of controlling a virtualized humanoid robot with 16 degrees of freedom (DOF). The aim of this study was to design an output feedback discontinuous controller which must resolve the sequence of articulation movements to walk over disjoint stepping objects placed in front of the humanoid robot. The suggested controller considers the state restrictions, via a nonstandard strong Lyapunov function, corresponding to the angular displacements and velocities at each articulation. This study implements an extended state terminal second order sliding mode controller with time dependent gains, to ensure the finite-time tracking trajectory of each articulation of the humanoid robot. The terminal sliding mode (TSM) is implemented in a virtual platform developed in a computer-aided design software. For comparison purposes, the controller was compared with a state feedback and a first order sliding mode algorithms.

This manuscript deals with the problem of controlling a virtualized humanoid robot with 16 degrees of freedom (DOF), each corresponding to the articulation in a real human being. That is, three DOF for each leg in the sagittal plane and one for the abduction movement; three DOF for each arm in the sagittal plane and one for the waist. The tracking trajectory problem of any humanoid robot like the classical biped robots requires a control algorithm with robustness against parametric uncertainties, fast response and even with finite-time convergence. These main characteristics are easily covered by sliding mode controllers. This manuscript implements a terminal second order sliding mode (TSOSM) controller to ensure the finite-time tracking trajectory of each articulation of the humanoid robot to the ones that define a classical walking pattern obtained by bio-mechanical studies. The TSOSM is implemented in a virtual platform developed in a computer-aided design software.

In this paper, a differential neural network (DNN) implemented as a robust observer estimates the dynamics of perturbed uncertain nonlinear systems affected by exogenous unknown inputs. In the first stage, the identification error converges into a neighborhood around the origin. Then, the second-order sliding mode supertwisting algorithm implemented as a robust exact differentiator reconstructed the unknown inputs. The approach proposed in this paper can be applied in the case of full access to the state vector (identification problem) and in the case of partial access to the state vector (estimation problem). In the second case, the nonlinear system under study must have well-defined full relative degree with respect to the unknown input. Numerical examples showed the effectiveness of the proposed algorithm. The first example tested the DNN working as an identifier into a mathematical model describing the dynamics of a spatial minisatellite. The second example (with a DNN implemented as an observer) tested the methodology of this paper over a single link flexible robot manipulator represented in a canonical (Brunovsky) form. In both examples, the mathematical models served as data generators in the testing of the neural networks. Even when not exact mathematical description of both models was used in the input estimation, the accuracy obtained with the DNN is comparable with the case of applying a high-order differentiator with complete knowledge of the plant.

The aim of this study was to develop a dynamic neural field (DNF) model to capture the essential non-linear characteristics of neural activity along several millimeters of visual cortex in response to local flashed stimuli. A two-layer DNF model was assessed to describe the response of both excitation and inhibitory layers of neurons. This particular structure of neurons interconnection was analyzed as a coupled system of non-linear integro-differential equations. This representation transformed the regular distributed form of DNF into an interconnected nonlinear model. A non-parametric modeling strategy yields to design the adjustment laws for the DNF weights. The algorithm used to adjust the weights considered self interconnections for each layer as well as external stimulus. The concept of controlled Lyapunov function served as the main tool to design a stable learning method for DNF. This algorithm was implemented in a class of hybrid computational model that served to execute the modeling of physiological response associated to visual external stimuli. The DNF model designed in this study can consider just the excitation response of specific neuron circuits without considering the presence of inhibitory response. This condition extends the number of electrophysiological trials where the adjusted DNF model can be evaluated. The learning method was evaluated with the information from a database that contains information coming from a selective visual attention experiment where the external stimuli appeared briefly in any of five squares arrayed horizontally above a central fixation cross. The degree of correlation (above 0.95) between signals measured at the brain cortex and the response of the DNF justified the application of the method proposed in this study.

This study addresses the problem of designing an output model reference control for non-linear systems in the presence of parametric disturbances/uncertainties in the state model and output noise measurements. A state observer based on a differential neural network (DNN) estimates the unknown states and the unknown disturbance simultaneously. The control design includes the estimated disturbance to provide a better tracking performance. The second result optimizes the gains of the controller and observer in order to obtain a reduced convergence zone for the tracking error based on the attractive ellipsoid method approach (AEM). Numerical results point out the advantages obtained by the nonlinear control based on the DNN observer when it is compared with a classical Luenberger structure.

The control problem of nonlinear systems affected by external perturbations and parametric uncertainties has attracted the attention for many researches. Artificial Neural Networks (ANN) constitutes an option for systems whose mathematical description is uncertain or partially unknown. In this paper, a Recurrent Neural Network (RNN) is designed to address the problems of identification and control of discrete-time nonlinear systems given by a gray box. The learning laws for the RNN are designed in terms of discrete-time Lyapunov stability. The control input is developed fulfilling the existence condition to establish a Quasi Sliding Regime. In means of Lyapunov stability, the identification and tracking errors are ultimately bounded in a neighborhood around zero. Numerical examples are presented to show the behavior of the RNN in the identification and control processes of a highly nonlinear discrete-time system, a Lorentz chaotic oscillator.

State estimation for uncertain systems affected by external noises is an important problem in control theory. This paper deals with the state observation problem when the dynamic model of a discrete plant contains uncertainties or is partially unknown. The suggested observer is oriented to solve the state observation problem of discrete time nonlinear systems. Most of the existing results using the neural networks approach have been developed for continuous time systems using. The recurrent neural network (RNN) have shown several advantages to treat many different control and state estimation problems. In this paper, it is presented a new discrete-time observer using the structure of a classical RNN. This observer includes a correction term using the output information and the first order sliding modes. The second method Lyapunov is applied to generate a new learning law, that contains an adaptive adjustment rate. This study proofs the stability condition for the free parameters included in the neural-observer. A numerical example is given using the RNN in the estimation of a mathematical model describing the HIV infection, that includes non-infected cells, infected cells and free virions. This model was used to generated the data using to test the observer.

State estimation for uncertain systems affected by external noises is an important problem in control theory. This paper deals with the state observation problem when the dynamic model of a plant contains uncertainties or is completely unknown and it is oriented to discrete time nonlinear systems because most of the existent results have been developed for continous time systems. The recurrent neural network (RNN) have shown his advantages to deal with this class problem. The Lyapunov second method is applied to generate a new learning law, containing an adaptive adjustment rate, implying the stability condition for the free parameters of the neural-observer. A numerical example is given using the RNN in the estimation of a mathematical model of HIV infection with three states.