Ameliorating Robot Execution Failures with Artificial Neural Networks: A Comprehensive Comparative Study

Abstract

The aim of this study is to investigate how ANNs can be trained on a personalized data set that contains several failure scenarios to predict and prevent execution errors in robotic systems. Three ANN architectures: ANN1 (single hidden layer), ANN2 (two hidden layers) and ANN3 (three hidden layers) were evaluated to determine their effectiveness in distinguishing between normal operating states and various fault conditions such as collisions and obstructions. The performance of the models is rigorously analyzed using several indices, including the average square error (MSE), precision, accuracy and F1 score. The results show that the ANN1 model has a simpler architecture and is more accurate in making predictions than more complex models. This shows that personalized machine learning approaches can make robotic systems safer and more reliable.