Softening the Links, Hardening the Math: An AI-Driven Solution for Forward Kinematics in Spherical Parallel Mechanism (SPM) with Soft Links

In this paper, we present an AI-driven approach to address the increasing complexity of forward kinematics in spherical parallel mechanisms (SPMs) with soft links. As robotic systems evolve toward more compliant and deformable structures, the challenge of accurately computing forward kinematics grows significantly due to the nonlinear deformation characteristics of soft materials. To address this challenge, we propose a data-driven solution that leverages machine learning techniques to model the soft link deformations and predict the configuration of the SPM efficiently. The proposed approach achieves accurate forward kinematics estimation without restricting the compliance of the soft links, preserving the safety and adaptability benefits they offer. We validate the approach using both analytically generated data from a rigid-link model and experimental data collected from a physical prototype, demonstrating that tree-based ensemble models, particularly Random Forest, achieve high orientation estimation accuracy across varying stiffness and loading conditions. This work represents a significant step toward making soft-link robotic systems more viable for complex applications, where precise positioning and adaptability are critical.