Personalized and safe gait pattern generation for lower extremity gait training using the Robot Assisted Training Platform (RATP)

Shamanth Shanmuga PrasadYoungwoo Kim^*

• Korea National University of Transportation, Chungju, North Chungcheong, Republic of Korea

Healthy humans exhibit diverse walking patterns. However, in clinical settings like hospitals and rehabilitation facilities, robot-assisted therapy often relies on pre-defined or reference gait trajectories, creating a one-size-fits-all approach. In contrast, the developed Robot Assisted Training Platform (RATP) distinguishes itself by generating personalized gait patterns tailored to each user's unique physical characteristics and rehabilitation goals. This study proposes a novel personalized gait pattern simulator that is used in the RATP for efficient lower extremity rehabilitation or training. The developed simulator integrates a personalized musculoskeletal model and simulation-based gait pattern generation using a Genetic Algorithm (GA). User-specific features such as musculoskeletal characteristics and physical state are incorporated into the model for personalization. The platform encompasses several key steps: kinematic and musculoskeletal model reconstruction, motion sequence generation, and forward and inverse dynamic analysis to synthesize natural walking motion. To achieve a gait pattern reflecting the user's gait training goals in rehabilitation application, a Genetic Algorithm optimizes muscle force generation across different muscle groups. However, for practical robotassisted training, the generated gait trajectory must consider both internal and external constraints. Internal constraints include joint range of motion, musculoskeletal characteristics, periodicity of the training trajectory, symmetricity between left and right leg trajectories, and walking speed. Whereas external constraints include stable training constraints based on ground contact condition of the feet and terrain information, ensuring the gait pattern is non-falling, smooth, and easy to follow, and maintains a limit cycle based on the gait transition model. These constraints are addressed in this study in the form of constraint equations or inequalities focusing on virtual space simulations. The gait patterns designed using this process are presented to the medical staff in the highest order of the evaluation function so that they could be chosen following a medical examination and applied to robot-