Session: Research Posters

Paper Number: 173900

Ai-Supported Optimization and Validation of Upper Limb Exoskeleton Design for Pediatric Rehabilitation 

Abstract

Introduction

Upper limb exoskeletons (ULEs) hold great promise for pediatric rehabilitation, where children face unique challenges such as rapid anatomical growth, frequent injuries, and the need for highly adaptable support systems. However, current commercial ULEs often fall short due to limited degrees of freedom (DOFs), restricted range of motion (ROM), and insufficient adaptability, which hinder their effectiveness in meeting the evolving needs of pediatric users.

This study explores the integration of AI-assisted design and simulation-based validation to address these limitations. Specifically, we propose a workflow that leverages machine learning and biomechanical simulation to optimize key exoskeleton specifications, such as DOFs and joint configuration, based on pediatric anatomical requirements. By systematically analyzing existing designs and refining new ones through AI-driven methods, we evaluate improvements in biomechanical performance and adaptability using dynamic simulation. A comparative analysis of AI-optimized 3D-printed prototypes and existing commercial devices will assess the extent to which this approach can overcome current design constraints. Ultimately, this work aims to demonstrate the potential of AI and simulation tools to enable more customizable, functional, and clinically relevant pediatric ULEs.

Contribution of the Work

This research advances ULE design by integrating AI-driven optimization and biomechanical simulation into a user-centered development process. The main contributions of this work include: (1) A simulation framework developed in MATLAB Simscape Multibody that incorporates pediatric biomechanical constraints to evaluate joint alignment, ROM, and structural performance under realistic load conditions. (2) An AI-assisted optimization pipeline that leverages simulation data to rapidly generate, refine, and evaluate exoskeleton designs tailored to individual patient needs. (3) A machine learning–based method for identifying key parameter relationships (e.g., DOFs, linkage lengths, joint placements), enabling interpretable, data-driven design decisions.

Methodology

The methodology consists of three primary phases: simulation-driven data generation, AI-assisted design optimization, and experimental validation. A set of ULE designs is modeled in SolidWorks based on pediatric anatomical requirements. These designs are imported into MATLAB Simscape Multibody, where biomechanical simulations are used to evaluate joint alignment, ROM, and structural integrity under pediatric load conditions. Simulation results are used to generate a structured dataset that links design parameters (e.g., linkage lengths, joint locations, DOFs) to performance metrics. This dataset forms the foundation for a surrogate modeling and optimization pipeline. An AI-driven design platform leveraging OpenAI guides the optimization process by evaluating and refining design variants through surrogate modeling, enabling rapid prediction of biomechanical performance and informed, patient-specific design decisions. The final design will be fabricated using 3D printing with polylactic acid (PLA), assembled, and evaluated in collaboration with an occupational therapist to assess usability, comfort, and functional performance in a rehabilitation context.

Preliminary Results and Conclusion

Preliminary work has focused on reverse engineering the Syrebo Upper Limb Dynamic Arm Support device (SY-UH01), a commercially available pediatric ULE, to identify core functional requirements such as joint configuration, assembly constraints, and ROM. Insights from this process informed the creation of an initial 3D model in SolidWorks, designed to meet performance objectives including quick assembly, three DOFs, and compatibility with common pediatric movement patterns. The model was then imported into MATLAB Simscape Multibody, where a simulation framework was established to evaluate joint torque and mechanical behavior under pediatric load conditions. This simulation environment also supports further development of control logic, including path-following for functional testing. The reverse engineering process additionally highlighted design features suitable for 3D printing, emphasizing rapid prototyping, lightweight construction, and cost-effective fabrication using PLA materials.

Future work will focus on generating a structured dataset from simulation outputs to support surrogate modeling and multi-objective optimization. This dataset will be used to uncover key parameter relationships and guide AI-assisted refinement of exoskeleton prototypes. Optimized designs will be fabricated and evaluated for functional performance, comfort, and adaptability in collaboration with clinical partners. The goal is to establish a data-driven framework for developing next-generation pediatric ULEs that are biomechanically sound, customizable, and clinically relevant.

Presenting Author: Jason Ly San Jose State University

Presenting Author Biography: Jason Ly is a Master's Student at San Jose State University, studying in the Mechanical Engineering Department with years of work experience in CNC programming and operation. Their research centers on integrating artificial intelligence and additive manufacturing to optimize user-centered exoskeleton solutions for pediatric healthcare. Jason has hands-on-experience with CAD modeling and simulation-based validation, with a recent focus on pediatric upper limb exoskeletons. He is passionate about bridging engineering, healthcare, and artificial intelligence to deliver accessible and effective rehabilitation tools.

Authors:

Yunjian Qiu San Jose State University
Jason Ly San Jose State University
Lin Jiang San Jose State University