Unified Neural Network-Based Shape Sensing for Continuum Robots Using Fiber Bragg Gratings

Continuum robots require precise shape sensing for reliable operation in constrained industrial and medical environments. We present a unified neural network that improves fiber Bragg grating shape sensing across two distinct continuum robots (a 165mm fully actuated robot and a 1.5m hybrid robot in a mock aero-engine) and multiple fiber configurations. Using camera-tracked ground truth, a single per-fiber model corrects distal positions from raw fiber Bragg grating estimates, implicitly compensating for twist and placement errors. On the 165mm robot we achieve 0.95mm root-mean-square error (0.57% length); on the 1.5m robot we achieve 2.02mm (0.13%) in 2-D and 1.78mm (0.11%) in 3-D, while reducing worst-case errors by up to an order of magnitude versus baselines. The model runs at >100 Hz and enables real-time, camera-free operation.