Inverse displacement analysis (IDA) is the determination of joints’ displacements in terms of the given position and orientation of a manipulator. Iterative and analytical methods are two common approaches for solving the IDA. Using analytical methods, the IDA problem is solved by applying trigonometry to generate closed-form solutions. Iterative methods can be used for solving the IDA when there is no closed-form solution, e.g., for redundant manipulators.Artificial neural network (ANN) is becoming an increasingly popular method for solving data related problems and has been applied in robotic topics. An ANN consists of neural layers containing artificial neural nodes. The neural nodes receive the input signals from the precedent neural layer, which mimics the operation of neurons in human brain. For data fitting, the entire dataset is divided into input dataset and output dataset. The input and output layers are used for recognizing the input and output data samples, respectively. The layers between the input layer and the output layer are called hidden layers. The connections among layers include weights, which measures the relative importance among signals for the specific node. The bias for each node shifts the preprocessed signal by a constant number. In default, weights and biases of the ANN are initialized randomly for ANN training. Backpropagation neural network is able to adjust the weights and biases between layers to minimize the mean squared error between the observed and the ANN identified output data. ANN is an effective tool for data regression. Transfer learning (TL) is to use the learned knowledge from the source task for the target task. ANN is one of the tools for achieving transfer learning.&nbsp;In this paper, an artificial neural network (ANN) based transfer learning approach of inverse displacement analysis of robot manipulators is studied. ANNs with different structures are applied utilizing data from different end effector paths of a manipulator for training purposes. Four transfer learning methods are proposed by applying pretrained initial parameters. Final training results of ANN with transfer learning are compared with those of ANN with random initialization. To investigate the rate of convergence of data fitting comprehensively, different values of performance targets are defined. The computing epochs and performance measures are compared. It is presented that, depending on the structure of ANN, the proposed transfer learning methods can accelerate the training process and achieve higher accuracy. For each method, the transfer learning improves the performance differently.

Neural Network Based Transfer Learning for Robot Path Generation

Paper Proceedings: https://asmedigitalcollection.asme.org/IDETC-CIE/IDETC-CIE2021/volume/85369

ASME 2021 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference (IDETC/CIE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.