Concurrent Learning Augmented DNN-Based and Admittance Control of a Hybrid Exoskeleton.

Area of Science:

• Robotics
• Neurorehabilitation
• Control Systems Engineering

Background:

• Neurological conditions (NCs) impair motor function, necessitating advanced rehabilitation strategies.
• Hybrid exoskeletons integrating functional electrical stimulation (FES) and robotic assistance offer promising rehabilitation solutions.
• Effective control of these hybrid systems requires managing both human muscle activation and robot motor dynamics.

Purpose of the Study:

• To develop and evaluate novel control strategies for hybrid exoskeletons.
• To enhance the performance and safety of FES-based rehabilitation equipment.
• To enable personalized and adaptive control for individuals with neurological conditions.

Main Methods:

• Development of two controllers: a deep neural network (DNN)-based controller and an Admittance-based controller.
• DNN used to approximate uncertain hybrid exoskeleton dynamics for efficient FES control, with multi-timescale weight updates (offline and online).
• Admittance-based controller utilizes torque feedback for adaptive motor control, prioritizing participant safety and comfort.

Main Results:

• The DNN-based controller demonstrated improved learning performance through a concurrent learning (CL) inspired term.
• The Admittance-based controller allowed for non-predetermined trajectories, enhancing participant interaction and safety.
• Lyapunov-based stability analysis confirmed the robustness and safety of both control systems.

Conclusions:

• The proposed DNN-based and Admittance-based controllers are effective for hybrid exoskeleton systems in neurological rehabilitation.
• These advanced control strategies improve system performance, adaptability, and user safety.
• The research contributes to the advancement of intelligent robotic systems for personalized physical therapy.