Transformer-based segmented learning for kinematics modelling of a cable-driven parallel continuum robot

Continuum robots with structural compliance exhibit significant operational potential in unstructured environments. However, theoretical kinematics modelling for continuum robots is complex due to their uncertain nonlinear characteristics. Although learning-based algorithms can effectively address the inherent challenges, unfortunately they frequently fail to capture internal information in robotic systems involving multiple spatial mappings, which reduces the accuracy of data-driven approaches in such applications. In this regard, we propose a segmented learning approach applied to kinematic modelling for a cable-driven parallel continuum robot (PCR), using Transformer networks to segmentally learn the mapping from task space to configuration space, and from configuration space to actuation space. Additionally, in order to evaluate the performance of the proposed approach, we employ multiple sets of neural-network models and various segmented learning configurations in the experiments. The experimental results across different trajectories show that, the Transformer network model under the segmented learning approach achieves superior trajectory tracking accuracy, compared to state-of-the-art modelling approaches.