Abstract:This study investigates modeling the dynamics of a 3D translational parallel manipulator with closed chains using feedforward neural networks (FFNNs). The dataset exceeds 50,000 samples, incorporating experimental data collected from a robot prototype using MATLAB® real-time workshop and the National InstrumentsTM DAQ toolbox, as well as CAD simulation data from MSC ADAMS software. While achieving satisfactory mean squared error (MSE), some predictions did not fully capture the manipulator's dynamics, with small overfitting observed. A Deep Neural Network (DNN) was tested but faced overfitting and high computational costs, despite being trained on a subset of the dataset. This highlighted the limitations of DNNs for modeling such complicated parallel robots with closed chains and parallelograms. FFNNs were preferred for their simplicity and lower overfitting risk. L2 regularization and k-fold validation were applied to improve performance. Transfer learning (TL) was also employed, fine-tuning a new network with weights from pre-trained FFNNs using a smaller, unseen dataset. This approach significantly reduced MSE and completely eliminated overfitting, demonstrating the effectiveness of TL in refining model performance for forward and inverse dynamics. These findings suggest that FFNNs, combined with TL, L2 regularization, and k-fold validation, offer a robust method for accurately modeling complex robotic dynamics, enhancing control and optimization strategies for complicated robotic systems. Training for all networks was conducted within the MATLAB® environment.

Augmented Neural Network for Full Robot Kinematic Modelling in SE(3)

A Robot Calibration Method Based on Joint Angle Division and an Artificial Neural Network

Simulation of a 7R Manipulator Inverse Calibration by Using an Optimized Neural Network Based on Beetle Antennae Search Algorithm

Modeling of robot inverse kinematics using two ANN paradigms

Learning complex nonlinear dynamics of a 3D translational parallel manipulator using neural network

High-Degrees-of-Freedom Dynamic Neural Fields for Robot Self-Modeling and Motion Planning

A hybrid method using FABRIK and custom ANN in solving inverse kinematic for generic serial robot manipulator

Kinematic and Nonparametric Calibration of a Parallel Machining Robot Based on an Artificial Neural Network

Dynamic Modeling of Robotic Manipulator Via an Augmented Deep Lagrangian Network

Learning-based Nonlinear Model Predictive Control of Articulated Soft Robots using Recurrent Neural Networks

A New Artificial Neural Network Approach in Solving Inverse Kinematics of Robotic Arm (Denso VP6242)

Comparison of numerical and neural network methods for the kinematic modeling of a hybrid structure robot

Autonomous Learning of the Robot Kinematic Model

Automating the Learning of Inverse Kinematics for Robotic Arms with Redundant DoFs

A calibration method for enhancing robot accuracy through integration of an extended Kalman filter algorithm and an artificial neural network

A neural network-based exploratory learning and motor planning system for co-robots

Data-Driven Inverse Kinematics Approximation of a Delta Robot with Stepper Motors

Comparing Semi-Parametric Model Learning Algorithms for Dynamic Model Estimation in Robotics

Knowledge-based Neural Ordinary Differential Equations for Cosserat Rod-based Soft Robots

Uncertainty-aware Active Learning of NeRF-based Object Models for Robot Manipulators using Visual and Re-orientation Actions