Model-Based Mid-Level Regulation for Assist-As-Needed Hierarchical Control of Wearable Robots: A Computational Study...

Area of Science:

• Robotics
• Human-Robot Interaction
• Control Systems Engineering

Background:

• Closed-loop human-robot systems necessitate controllers that model both human and robot dynamics.
• Human adaptation to robotic systems is crucial for effective control and interaction.
• Assist-as-needed (AAN) policies aim to provide compensatory torque for robot dynamics and external loads.

Purpose of the Study:

• To develop and evaluate novel mid-level controllers for AAN policies in a hierarchical control setting.
• To compare model-based and fuzzy logic approaches against traditional methods for AAN.
• To simulate and analyze human adaptation to robotic systems under varying experience levels.

Main Methods:

• Implementation of a hierarchical control structure with a mid-level controller for AAN policies.
• Development of two novel AAN methods: model-based control and fuzzy logic rule-based control.
• Simulation of human adaptation using a nonlinear model predictive controller (NMPC) representing the human central nervous system (CNS) across initial, short-term, and long-term experiences.

Main Results:

• Both model-based and fuzzy logic AAN methods demonstrated superior performance compared to the traditional proportional method.
• The CNS NMPC exhibited challenges in initial experiences, activating antagonist and agonist muscles to mitigate oscillations.
• Long-term adaptation simulations showed no oscillations as the CNS NMPC learned the robot model and adjusted its parameters.

Conclusions:

• Gradual increase in robot strength is recommended to manage unexpected human-robot interactions like spasticity or vibration.
• The proposed mid-level controllers are suitable for applications in wearable assistive devices, exoskeletons, and rehabilitation robots.
• Effective human-robot interaction relies on controllers that account for human adaptation and system dynamics.