Workspace Identification Using Neural Network for an Optimal Designed 2-DOF Orientation Parallel Device

Tănase, Ionela; Itul, T.; Campean, Emilia; Pîslă, Adrian

doi:10.1007/978-94-007-4902-3_17

Cited by 3 publications

(4 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Neurons in the first layer receive external inputs as follows: 24) provides the starting point for (23). The neural network shown in Figure 6 consists of 2 inputs, three hidden layers with 15 neurons per layer, and an output layer of 8 neurons and 8 outputs, according to (22). This network has associated in the neurons of the hidden layer transfer functions (also called activation function) hyperbolic tangent sigmoid and in the neurons of the output layer linear transfer functions, and so, maintains the properties of nonlinearity and linearity between the input and output data of the network.…”

Section: Neural Network Architecturementioning

confidence: 99%

“…Among the algorithms available to train neural networks are several high-performance algorithms that can converge ten to one hundred times faster than other algorithms, such as the Levenberg-Marquardt algorithm which is considered the fastest method to train moderately sized feedforward neural networks (up to several hundred weights) [55]; it also has an efficient implementation in the MATLAB software ® . For neural network training, this work used the Levenberg-Marquardt algorithm; the data used are those described in (22). The dataset was not divided for testing and validation because the network is intended to fully learn all the data from the area (AT)) portion considered; the target mse that the network must achieve in its performance was set to 1e -6 and the maximum number of epochs was 3000.…”

Section: Neural Network Trainingmentioning

confidence: 99%

“…Artificial neural networks represent another alternative to solve the inverse problem of serial and parallel robots [21]. For example, in [22] neural networks are used to generate the workspace of a 2-DOF parallel mechanism by taking information from the solution of the reverse kinematic problem. In [23] three types of neural networks (multilayer perceptron, radial basis function, and Local Linear Model Trees) are used to solve inverse and direct kinematics, This work is licensed under a Creative Commons Attribution 4.0 License.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Modeling in Two Configurations of a 5R 2-DoF Planar Parallel Mechanism and Solution to the Inverse Kinematic Modeling Using Artificial Neural Network

López

Mora-Pulido

et al. 2021

IEEE Access

View full text Add to dashboard Cite

This article introduces a new kinematic modeling method used to analyze coupled rigid multibody movements. The method was applied to the study of a 5R planar parallel mechanism's kinematics and consists of analyzing two fixed configurations of the mechanism to systematize the rotational relationships between the two structures. Mathematical models were developed using complex numbers. The inverse kinematic problem was modeled as a system of eight nonlinear equations and eight unknowns, which was solved with Newton-Raphson's method. Subsequently, with the inverse problem model, a numerical database related to the mechanism configurations, including singular positions, was generated to train a multilayer neural network. The Levenberg-Marquardt algorithm was used for network training. Finally, an interpolated linear path was used to understand the efficiency of the trained network.

show abstract

Section: Neural Network Architecturementioning

confidence: 99%

Section: Neural Network Trainingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Modeling in Two Configurations of a 5R 2-DoF Planar Parallel Mechanism and Solution to the Inverse Kinematic Modeling Using Artificial Neural Network

López

Mora-Pulido

et al. 2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The result is repeated for 10 training sessions, each with 10 3 epochs done on the training set. Figure (6) presents the prediction error of an independent test set of 10 3 samples evaluated on the best performing session after training. With a filter function of threshold value 0.5, the network is able to predict over 99% of the samples correctly.…”

Section: Experimental Setup: 6-dof Serial-link Manipulator With Spher...mentioning

confidence: 99%

Deep Neural Network Based Subspace Learning of Robotic Manipulator Workspace Mapping

Liao

2018

2018 International Conference on Control, Artificial Intelligence, Robotics &Amp; Optimization (ICCAIRO)

View full text Add to dashboard Cite

The manipulator workspace mapping is an important problem in robotics and has attracted significant attention in the community. However, most of the pre-existing algorithms have expensive time complexity due to the reliance on sophisticated kinematic equations. To solve this problem, this paper introduces subspace learning (SL), a variant of subspace embedding, where a set of robot and scope parameters is mapped to the corresponding workspace by a deep neural network (DNN). Trained on a large dataset of around 6 × 10 4 samples obtained from a MATLAB R implementation of a classical method and sampling of designed uniform distributions, the experiments demonstrate that the embedding significantly reduces run-time from 5.23 × 10 3 s of traditional discretization method to 0.224 s, with high accuracies (average F-measure is 0.9665 with batch gradient descent and resilient backpropagation). 1

show abstract

Full Workspace Generation of Serial-link Manipulators by Deep Learning based Jacobian Estimation

Liao¹,

Mao²

2018

Preprint

View full text Add to dashboard Cite

Apart from solving complicated problems that require a certain level of intelligence, fine-tuned deep neural networks can also create fast algorithms for slow, numerical tasks. In this paper, we introduce an improved version of [1]'s work, a fast, deep-learning framework capable of generating the full workspace of serial-link manipulators. The architecture consists of two neural networks: an estimation net that approximates the manipulator Jacobian, and a confidence net that measures the confidence of the approximation. We also introduce M3 (Manipulability Maps of Manipulators), a MATLAB robotics library based on [2](RTB), the datasets generated by which are used by this work. Results have shown that not only are the neural networks significantly faster than numerical inverse kinematics, it also offers superior accuracy when compared to other machine learning alternatives. Implementations of the algorithm (based on Keras [3]), including benchmark evaluation script, are available at https://github.com/liaopeiyuan/Jacobian-Estimation. The M3 Library APIs and datasets are also available at https://github. com/liaopeiyuan/M3.

show abstract

Workspace Identification Using Neural Network for an Optimal Designed 2-DOF Orientation Parallel Device

Cited by 3 publications

References 7 publications

Modeling in Two Configurations of a 5R 2-DoF Planar Parallel Mechanism and Solution to the Inverse Kinematic Modeling Using Artificial Neural Network

Modeling in Two Configurations of a 5R 2-DoF Planar Parallel Mechanism and Solution to the Inverse Kinematic Modeling Using Artificial Neural Network

Deep Neural Network Based Subspace Learning of Robotic Manipulator Workspace Mapping

Full Workspace Generation of Serial-link Manipulators by Deep Learning based Jacobian Estimation

Contact Info

Product

Resources

About