Muhammad Majid Gulzar

Comparison of GPGPU based robotic manipulator with other embedded controllers

Industrial and robotic controllers have to execute various complex independent tasks repeatedly in real time. In order to implement these algorithms with non-linear equations, massive computational power is required in a motion control system. In this paper, inverse kinematics algorithm is selected as a test algorithm to compare performance of General Purpose Graphics Processing Unit (GPGPU) with other widely used sequential and concurrent controllers. Inverse Kinematics algorithm is implemented sequentially in Arduino Due microcontroller and FPGA is used for concurrent implementation where algorithm is designed in VHDL using combinational division. Execution speeds of these controllers are compared with NVIDIA Quadro K2200 GPU programmed with CUDA Parallel Computing Architecture. Results validated that using computational power of GPU, execution time of large independent tasks is significantly decreased.

Towards MOOCs and Their Role in Engineering Education

Massive Open Online Courses (MOOCs) are the latest installment in the field of distance education. This paper discusses the pros and cons of MOOCs in the educational systems with a special emphasis on the engineering education. Since 2012, more and more universities have been joining MOOC revolution and the number of online courses has considerably increased over time. Engineering MOOCs are also increasing. Due to their access to large audience, massive online courses have been expanding their horizon of admission to engineering education at all levels and improving the in-campus learning.

Realization of an improved path planning strategy

This paper presents an improved path planning algorithm to avoid obstacles of critical shapes (obstacles having angles of around 90°). The proposed algorithm outperforms the reported bug algorithms by reaching the target safely in relatively less time and avoiding multiple obstacles without suffering into local minima. The main reason of such improvement is to schedule the speed and direction of the agent according to the edges of obstacles. More specifically, the agents speed is set as a function of goal and the obstacle location. The direction to follow obstacle edges is not predefined, but generated online by taking the dot product between obstacle edges and the target behavior. In this way the optimal path can be achieved to reach the final target with safe, smooth and short trajectory. Simulation results and practical implementation validated the effectiveness of our proposed algorithm. Moreover, performance comparison of the proposed algorithm and other approaches confirms the efficiency of the presented approach.

Implementation of image processing based Digital Dactylology Converser for deaf-mute persons

Language of signs is usually used by the deaf-mute persons in order to convey their message. But, it becomes very difficult for those who have no knowledge of the sign language. The aim of this research work is to resolve this problem using Digital Dactylology Converser (DDC) which is a device that converts a sign language into voice signal and text message. Primarily a hands gesture is recognised by computer vision program and based on that image a speech and text is played and displayed on the speaker and LCD respectively. Purpose of this device is to make conversation easy between dumb and the normal people around the world. In order to minimize their limitations and enable them to expand their boundaries, this is best suitable device. Experimental results verify the operational accuracy of the device for different users. The proposed device can work well and translates the letters, alphabets to text and sound.

Kinematic modeling and simulation of an economical SCARA manipulator by Pro-E and verification using MATLAB/Simulink

SCARA (Selective Compliance Assembly Robot Arm) manipulators are among the most extensively used manipulators in industry due to high precision and their inherent rigidity. This research demonstrates mechanical design process of an economical SCARA manipulator with unique and competing industrial specifications using Pro-E software. Inverse kinematic equations are also derived using algebraic and geometric method to control the manipulator movement. The design process includes the design of joints, links and controller as well as the selection of its electrical and mechanical components. The selection of the actuators and the dimensions of the whole mechanical structure are selected in such a way to direct its center of gravity towards base and also reduce the vibration and backlash in its mechanical structure. The main task was to use readily available components with an eye on keeping the costs down. Finally the performance of SCARA system is examined in Pro-E and verified manipulators movement with MATLAB/Simulink which exhibits the effectiveness of the proposed model.

Implementation of electroencephalography controlled prosthetic hand

The purpose of brain machine interface (BMI) is to provide a communication path between brain signals and external devices. Use of brain signals for restoring function of impaired body parts using scientific methods require brain machine interface. This article presents BMI to control five-fingered prosthetic hand using the electroencephalography (EEG) signals. The core objective of the system is to develop a control system that will be able to communicate with the brain thoughts. EEG signals captured from the human scalp are used as information carrier for brain control interface (BCI) system. This research article discusses the development of a system to assist disable persons using EEG based signals.

GPGPU based concurrent classification using trained model of handwritten digits

In this paper, General Purpose Graphical Processing Unit (GPGPU) based concurrent implementation of handwritten digit classifier is presented. Different styles of handwriting make it difficult to recognize a pattern but using neural network, it is not a difficult task to perform. Different softwares like torch and MATLAB provide the support of multiple training algorithms to train a network. By choosing an appropriate training algorithm for a specific application, speed of training can be increased. Furthermore, using computational power of GPUs, training and classification speed of neural network can be significantly improved. In this work, Modified National Institute of Standards and Technology (MNIST) database of handwritten digits is used to train the network. Accuracy and training time of digit classifier is evaluated for different algorithms and then concurrent training is performed by exploiting power of GPU. Trained parameters are imported and used for the concurrent classification with Compute Unified Device Architecture (CUDA) computing language which can be useful in numerous practical applications. Finally, the results of sequential and concurrent operations of training and classification are compared.

Dynamics Study of Compass-Gait Biped Robot through Nonlinear Time-Series Analysis

Passive dynamic walking (PDW) provides us better insight for understanding human walking, for developing prosthetic limbs and for designing superior bipedal robots. In this paper, we investigated the dynamics of a simple PDW, 2D compass-gait biped model that loosely look like human legs (without knees), using time-series analysis based on nonlinear dynamics. Previously, this passive biped model has been explored using only bifurcation diagrams and phase-space plots, but we studied it using nonlinear time-series analysis. The experimental results from walking experiments of prototype passive compass-gait biped validated the simulation results. These results can be useful for designing efficient bipedal robots.

Design and implementation of a single-phase multilevel cascaded inverter for PV system

Electrical energy generation by using Photovoltaic system is becoming popular day by day. This paper presents a control topology for multilevel inverters used in standalone PV system. A hybrid technique is used to optimize the switching angles of thirteen levels single-phase cascaded multilevel inverter for selective harmonics elimination. The nonlinear transcendental equations problem is solved by dividing the whole system into two stages. In stage-I: it will find the convergence angles region by using Fmincon optimization technique and then in stage II it will find the accurate convergence point by using most common technique which is Newton Raphson (NR) method. This approach is fast, low cost, efficient and it produces very low total harmonics distortion. The ability of the proposed technique is simulated by using MATLAB and implemented on cascaded h-bridge circuits, opto-isolators and Arduino UNO microcontroller. The simulation and hardware results verify the capability of proposed technique.

A hybrid technique to harvest maximum power from PV systems under partial shading conditions

Maximum Power Point Tracking (MPPT) under partial shading condition is the main issue in Photovoltaic (PV) system. This research work is based on a hybrid mechanism for MPPT problem by utilizing two commonly used techniques. The whole algorithm is divided into two stages. Initially it uses Fractional Open Circuit Voltage (FOCV) technique in order to detect Global Maxima (GM) region. Once the GM region is achieved the variable step sized incremental conductance technique can easily detect the Maximum Power Point (MPP). The proposed technique is fast, accurate and simple to implement. This algorithm is implemented and tested using MATLAB/Simulink and the robustness of this algorithm is verified by testing this algorithm under different partial shading conditions. Finally the superiority of this algorithm is judged by comparing this algorithm with other MPPT algorithms.