Juniper Publishers| Robotics Engineering

Neural Network Training Using Unscented and Extended Kalman Filter




Authored by Denis Pereira de Lima,

This work demonstrates the training of a multilayered neural network (MNN) using the Kalman filter variations. Kalman filters estimate the weights of a neural network, considering the weights as a dynamic and upgradable system. The Extended Kalman Filter (EKF) is a tool that has been used by many authors for the training of Neural Networks (NN) over the years. However, this filter has some inherent drawbacks such as instability caused by initial conditions due to linearization and costly calculation of the Jacobian matrices. Therefore, the Unscented Kalman Filter variant has been demonstrated by several authors to be superior in terms of convergence, speed and accuracy when compared to the EKF. Training using this algorithm tends to become more precise compared to EKF, due to its linear transformation technique known as the Unscented Transform (UT). The results presented in this study validate the efficiency and accuracy of the Kalman filter variants reported by other authors, with superior performance in non-linear systems when compared with traditional methods. Such approaches are still little explored in the literature

For more open access journals in Juniper Publishers please click on: https://juniperpublishers.com

For more open access journals on Journal of Robotics & Automation Engineering please click on https://juniperpublishers.com/raej/

To read more please click on: https://juniperpublishers.com/raej/RAEJ.MS.ID.555568.php


Comments

Post a Comment

Popular posts from this blog

Geometric Simulation for 3D-Printed Soft Robots- Juniper Publishers

Image
Juniper Publishers- Journal of Robotics Introduction Robots fabricated by soft materials can provide higher flexibility and thus better safety while interacting with natural objects with low stiffness such as food and human beings. With the growth of three-dimensional (3D) printing, it is even possible to directly fabricate soft robots [1,2] with complex structures and multiple materials realizing highly dexterous tasks like human-interactive grasping and confined area detection [3]. However, with such increased degrees of freedom (DoF), the design of soft robots becomes a very difficult task. It can be made possible by integrating simulation into the design phase. However, the shape deformation comes from many different and complex factors including manufacturing process, material properties, actuation, etc. Especially with the limited understanding of layer-based additive process in AM, it is challenging to formulate a complete mathematical model for the simul
Read more

Industry 4.0`S Product Development Process Aided by Robotic Process Automation- Juniper Publishers

Image
Juniper Publishers- Journal of Robotics Opinion As well defined on the journal`s site Robotic Process Automation (RPA) is the application of technology that allows employees in a company to configure computer software or a “robot” to capture and illustrate existing applications for processing a transaction, managing data, provoke responses and communicating with other digital systems. Product Development Process (PDA) is a key element for implementing Industry 4.0 and as defined in literature it consists of integrating six main elements: requirement, definition, design, production, maintenance and recycling [1-3]. So, in order to implement Industry 4.0 fast and costless we need to capture and illustrate existing applications as well as new ones for processing ensuring a reliable product-life cycle [4] and to efficiently process transactions, assuring smart services, manage big data and communicate with other digital systems. In other words Industry 4.0 is r
Read more

Research on Electromagnetic Driving Robot Fish-Juniper Publishers

Image
Juniper Publishers-Journal of Robotics Abstract Based on the actual physical appearance of the tuna and the wave equation derived from observing its swimming locomotion, this paper proposes a new idea on the simulation and modeling of the robot fish and a novel way of thinking on kinetic analysis. The motion law for each joint of the robot fish was first obtained through the discrete fitting method and the motion law was subsequently used as the basis for the robot fish electromagnetic drive control signal. The fluent flow field analysis software was utilized. The meshing of the fluid around the fish body was performed by using the dynamic grid technique. The surface pressure values of the fish body during the steady state forward movement were analyzed, and the values of the driving force of each part of the fish were obtained. It was then determined that the electromagnetic drive caudal fin robot fish was the optimal design. Based on the idea of digital-analog c
Read more