Strategy with machine learning models for precise assembly using programming by demonstration

Programming by demonstration (PBD), which is applied in industry as a method of human–robot collaboration for assembly tasks, such as placing, inserting, and screwing, has seen limited application in commercial electronic product assembly due to a lack of trajectory planning optimization. In this research, we propose a framework with two custom algorithms to preprocess and classify contactless demonstration performance. This framework enables the generation of optimal motion paths based on that criteria of distance, smoothness, and trajectory variance rather than canonical methods. Machine learning methods, including convolutional neural network (CNN), artificial neural network (ANN), and support vector machine (SVM), are feasible for predicting the best motion path with an accuracy range of 80% to 85%. Among these methods, CNN, specifically DarkNet, achieves the highest accuracy. Future work involves development of hybrid CNN/ANN algorithms, which may yield higher accuracy in prediction. In addition, the proposed algorithms may be applied to robots equipped with dual assembly arms and other complex assemblies that emulate human arms.