Forward geometric model prediction of a 6-RSU parallel manipulator using a modified NARX Bayesian neural network

Area of Science:

• Robotics
• Machine Learning
• Control Systems

Background:

• Accurate robot models are crucial for precision and safety in robotic applications.
• Bayesian Neural Networks (BNNs) can model complex systems and quantify uncertainty.
• Existing machine learning methods for Forward Geometric Model (FGM) prediction often lack uncertainty estimation.

Purpose of the Study:

• To propose and evaluate a modified Nonlinear Autoregressive with Exogenous Inputs - Bayesian Neural Network (NARX-BNN) for enhanced FGM prediction.
• To improve the accuracy and reduce predictive uncertainty in FGM estimation for a 6-RSU parallel manipulator.
• To demonstrate the superiority of the proposed NARX-BNN over traditional BNNs.

Main Methods:

• Development of a modified NARX-BNN model integrating BNNs' approximation and uncertainty capabilities with nonlinear ARX predictive power.
• Application of the NARX-BNN to predict the FGM of a 6-RSU parallel manipulator.
• Comparison of NARX-BNN performance against traditional Bayesian shallow neural networks using Variational Inference.

Main Results:

• The NARX-BNN model demonstrated superior performance compared to traditional BNNs.
• At a 95% confidence level, NARX-BNN reduced the Root Mean Square Error (RMSE) by up to 11%.
• NARX-BNN decreased the Average Width of the prediction interval by approximately 12.7%.

Conclusions:

• NARX Bayesian Neural Networks significantly enhance FGM prediction accuracy and reduce uncertainty in robotic applications.
• The proposed method improves the reliability of machine learning models for parallel manipulators.
• Advancements hold promise for better robotic control, planning, and overall system dependability.