Multi-Modal Muscle Activation Modeling Using Koopman Operator Linearization for an Ankle Exoskeleton

Gait impairments impact mobility and quality of life, necessitating intelligent rehabilitation devices that utilize dynamic models to improve the human-machine interaction and provide effective assistance. This study presents a data-driven model that integrates Koopman operator linear estimations with fused surface electromyography (sEMG) and ultrasound (US) imaging for estimating volitional ankle torque throughout the gait cycle. The Koopman operator estimates the nonlinear, phase-dependent dynamics of ankle motion, while the fusion of sEMG and US signals enhances muscle activation predictions by combining electrical and morphological insights. This framework demonstrates the potential to improve state and volitional torque estimation accuracy, laying the groundwork for individualized exoskeleton assistance and enhanced gait rehabilitation strategies.