Search

Riemannian Motion Policies (RMPs) have recently been introduced as a way to specify second-order motion policies defined on robot task spaces. RMP-based approaches have the advantage of being more general than traditional approaches based on operational space control; for example, the generalized task inertia in an RMP can be fully state-dependent, which is particularly effective in designing collision avoidance bahaviors. But until now RMPs have been applied only to fully actuated systems, i.e. systems for which each degree of freedom (DoF) can be directly actuated by a control input. In this paper, we present a method that extends the RMP formalism to a class of underacutated systems whose dynamics are amenable to a decomposition into a fully-actuated subsystem and a residual dynamics. We show the efficacy of the approach by constructing a suitable decomposition for a Wheeled-Inverted-Pendulum (WIP) humanoid robot and applying our method to derive motion policies for combined locomotion and manipulation tasks. Simulation results are presented for a 7-DoF system with one degree of underactuation.

Microgrids are becoming very popular nowadays throughout the world as they utilize renewable energy resources (RER) effectively. Due to the intermittent nature of the dominant RER, different kind of energy storage systems are introduced to enhance reliability, stability, and performance of the microgrids. This paper presents an effective control and energy management strategy of a direct current (DC) microgrid employing hybrid energy storage systems (HESS). The main components of the modeled microgrid are intermittent RER, variable load, HESS, backup diesel generator, and power electronic converters. The current-mode controller manages to charge and discharge the HESS to overcome the intermittency of the RER and load variation. Besides, it proposes an effective energy management strategy that compensate for the energy generation shortage of the RER by utilizing the backup diesel generator. Furthermore, it maximizes the power generation from the photovoltaic (PV) plant employing maximum power point tracker (MPPT). Finally, this research verifies the effectiveness of the proposed system through simulation where the DC voltage level never crosses ± 5% limit under load variation and RER intermittency.

Text summarization is an approach by which the size of one or more document is shortened and the shorten passage presents the core information of the document. In this modern era of information technology, we are over flooded with online data which raised the necessity of summary of the original text. Many methods have already implemented for English text and the effort for Bengali text are gaining alongside. In this paper, we propose an extractive text summarization technique based on a deep learning model of Recurrent Neural Network (RNN) for single document summary. Our method is to classify the sentences as significant or not for the summary. We have used Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU) based RNN. Between them, we found LSTM more promising and we achieved average F1 scores- 0.63, 0.59, 0.56 for Rouge-1, Rouge-2 and Rouge-3 in some respects.

This study takes up a topic in wave generation. In a wave tank of hidrodynamic laboratory well known two types of wavemakers, i.e. flap-type and piston type, are used to generate waves. In addition to these two types of wavemakers, there is another type of wavemaker which is the snake-type. Snake-type wavemakers consists of many parts which can move like flaps or pistons. Here, we investigate the propagation of waves generated by the snake-type wavemakers, which is flap type, based on the wave propagation governing equation (Laplace equation). Based on the linear theory we derived a ratio between the stroke maximum of wavemakers (S) and the amplitude (A) of generated waves. It is found that the ratio of S and A does not only depend on frequency of generated wave but also depend on water depth. The formulation for the wavemaker control will also be analyzed and simulated.

We present a method to implement operational space control technique for whole body control of a wheeled-inverted pendulum (WIP) humanoid, a class of robots that feature a redundant manipulator dynamically balancing itself on two wheels. We consider the case in which the base link is not passive, in that, the reaction of the torque driving the wheels is acting on the base link. This makes the system underactuated because the number of actuators is less than the degrees of freedom. However, it is a special form of underactuation where all degrees of freedom are actuated, with one actuator being shared by two degrees of freedom. This special feature of WIP dynamics challenges the straightforward application of operational space techniques, compared to fully actuated, constrained, or underactuated systems with no DOFs sharing the actuators. The key idea is to isolate dynamics of the manipulator from the wheel dynamics and to control the resulting system using operational-space control. A side-effect of this is the resulting asymmetry of the inertia matrix restricting full use of operational space tools. This is addressed by presenting a coordinate transformation that transforms the inertia matrix into a symmetric one. Finally, the wheel dynamics is derived as a function of manipulator joint accelerations so that horizontal motion of the robot can be treated as another task being controlled within the framework developed for manipulator control. Simulation results are presented on a five degree of freedom planar robot, with four DOF serial arm mounted on wheels.

This paper concerns on an optimal control problem of a robot movement from an initial point to another target point while avoiding obstacles. Previous study had generated a trajectory that should be taken by the robot to minimized the consumed energy while moving with constant velocity. Here, we find the optimal performance index value which minimizes the trajectory length as well as the energy usage of the robot. We also consider nonconstant velocity which changes over time as the robot moves. The solution is generated numerically by using software package TOMLAB/PROPT. The results show that there is different route taken by the robot yet it is shorter in distance. Also, the optimum energy and trajectory length can be obtained by setting the right penalty values at both measures.

Partitioning vast amounts of text documents is a challenging problem due to a high dimensional representation of the documents. In this study, we investigate the quality of text document clustering when Singular Value Decomposition (SVD) is used to reduce the dimension of the documents. The results show that the quality of the clusters is very comparable to that of when the dimensions are not reduced. In addition, the computational cost to cluster documents can be reduced significantly when the clustering is done on a small dimension.

This paper presents the novel idea for opportunity detection (detecting unoccupied sub-carriers/sub-channels) in downlink (DL) orthogonal frequency division multiple access (OFDMA) systems using cognitive radio (CR) sensing techniques and sub-channel utilization history information. Assuming availability of information about the basic parameters of the primary system as well as time and frequency synchronization of CR user with primary system, three techniques have been proposed for detection of unoccupied sub-channels in DL OFDMA systems: 1) Detection Method— it identifies unoccupied sub-carriers/sub-channel in DL OFDMA systems by exploiting the CR sensing techniques (i.e. matched filter and energy detection). 2) Prediction Method— it forecasts the future sub-channel state (empty or occupied) by classifying sub-channel history into different traffic patterns using ‘Traffic Classification Algorithm’. 3) Hybrid of Detection and Prediction Method— it also utilizes ‘Traffic Classification Algorithm’ to decide when to use detection and prediction methods in order to determine either sub-channel will remain empty or occupied in next time slot.

In this paper, a MIMO nonlinear discrete time model of the idle mode of an SI engine is first developed using a fuzzy logic model in the form of a nonlinear radial basis function expansion. The inputs are the throttle angle and the ignition advance and the outputs are the idle speed and manifold pressure. The model is validated using nonlinear correlation tests for MIMO systems. Phase portrait analysis is used to study the global dynamic behavior of the plant, in order to find periodic motions and equilibrium points of the system. Optimal state trajectories are used as entries to a fuzzy logic expansion to approximate the distorted control surface of the plant. The resulting FAM or fuzzy controller smoothens the control surface, and consequently, the transitions of the state trajectories from a different initial condition to another in its path to the equilibrium. Both the model and controller parameters are optimized using a modern version of an evolutionary algorithm, called covariance matrix adaptation-evolution strategies (CMA-ES).

Modeling of nonlinear systems is implemented using a partial linear modeling (PLM) technique, which separates the linear and nonlinear part of the model. The linear part of the model is in the form of an ARX model, while the nonlinear part is in the form of NNARX model using an RBF neural network. For the RBF neural network, the centers of the network are chosen using an orthogonal least squares (OLS) method. The linear part of the model is constructed to fit and absorb as much as possible the dynamic of the system, while its residuals are fitted using the nonlinear part of the model. The model is tested on experimental data of a MIMO spark ignition (SI) engine system. The plant (SI engine) is handled as a two-inputs, two-outputs process, the two inputs are the ignition timing and the throttle angle, the two outputs are the engine speed and manifold pressure. Different order and nonlinear terms of model are tested on the input-output data to obtain a valid model. Finally, second order with three nonlinear terms of model is found as a fairly accurate model. Model validation is treated using different sets of input-output data which indicates that the resulting model is fairly valid. A feedback linearization method is used for controlling the nonlinear system where the model is constructed using the PLM technique. The smooth transition of the controller output shows that the combination of the modeling and control technique has real potential for real-time implementation.