Enhancing Machine Learning-Based Feedforward Control of 2-DOF Flexure Manipulator: Benefits of Time-Delay Embedding

This research uses machine learning techniques to enhance a feedforward controller for a fully actuated 2 degrees of freedom manipulator with flexure joints. The foundation of the controller is a combination of the Lagrangian Neural Network to model the system’s conservative forces and the Feedforward Neural Network to simulate non-conservative ones. To address the limitations of both networks in precisely modeling the reproducible part of these forces, we introduce the weighted least-squares method with regularization, which maps the system’s configurations to the residue of control signals (error) and adjusts the model with rank-1 updates. Inevitable trade-offs apply when one uses Time-Delay Embedding, but the preliminary results indicate its feasibility in application to improve the used error learning approach.