Geometry-Adaptive Kinematics and Experimental Validation of a Linear-Actuator-Driven Parallel Lumbar Exoskeleton

Abstract

Wearable lumbar rehabilitation exoskeletons require body-size adaptability and deterministic kinematic modeling. Adjustable wearing structures may alter actuator anchor-point positions, creating inconsistencies in fixed-geometry inverse kinematic models. This study proposes a linear-actuator-driven parallel lumbar exoskeleton using a split semi-ring variable-width platform. Transverse opening increments, h1 and h2, are introduced into the anchor-point coordinates to reconstruct the worn geometry. A geometry-adaptive kinematic model is developed to map task-space motions to actuator-space commands. Multibody simulations indicate that compared with a fixed-geometry model, the proposed approach decreases the position RMSE from 1.30mm to 0.32mm and the orientation RMSE from 0.62∘ to 0.05∘. Human-worn experiments across three healthy adult male participants yielded a global actuator displacement RMSE of 0.08–0.09mm, a task-space position RMSE of 4.73–7.64mm, and an orientation RMSE of 0.71∘–1.09∘. These task-space errors characterize the exoskeleton-platform motion measured by optical marker clusters rather than directly measured anatomical lumbar-spine kinematics. Rigid testbench validation isolated the kinematic mechanism from human-interface compliance, reducing the testbench position RMSE from 3.07mm to 1.86mm and the orientation RMSE from 0.78∘ to 0.34∘ compared to the fixed-geometry assumption. These results provide a proof-of-concept validation for the feasibility of the architecture and model for lumbar exoskeletons with transverse width adjustability.