Soft origami robots offer unique advantages in flexibility and adaptability. However, their precise inverse kinematics modeling remains a challenge due to highly nonlinear kinematic behavior caused by actuator backlash and material compliance. This work proposes an approach based on an Adaptive Neuro-Fuzzy Inference System (ANFIS) to approximate the inverse kinematics of a Yoshimura origami-based continuum robot. A camera vision system is used to acquire robot motion data, using the pinhole model and Zhang’s method for camera calibration. Reliable data (x, y end-effector coordinates) for kinematic learning were obtained using the calibrated system. The ANFIS system converged after 150 training epochs and was experimentally validated using quantitative error metrics, including repeatability (0.009-0.190 cm) and accuracy (0.052-0.355 cm), despite nonlinearities associated with TPU-95A hyperelastic behavior. Unlike conventional model-based approaches, the proposed ANFIS learning framework provides an experimentally validated data-driven solution to the inverse kinematics problem of tendon-driven continuum robots and demonstrates robustness within the validated operating region under tendon backlash and material compliance.
23 June 2026 in Iranian Journal of Science and Technology, Transactions of Mechanical Engineering