Pilot Testing of Dynamic Bayesian Optimization for Exoskeleton-Assisted Training of Propulsion Mechanics.

Area of Science:

• Biomechanics
• Robotics
• Neuroscience

Background:

• Human-in-the-loop (HIL) optimization is effective for assistive tasks but not yet for training.
• Conventional HIL methods struggle to incorporate neuromotor learning during training.
• Neuromotor learning is crucial for adapting movement patterns, such as gait mechanics.

Purpose of the Study:

• To implement and evaluate dynamic Bayesian optimization (DBO) for HIL training of gait propulsion mechanics.
• To investigate the effect of DBO on increasing the trailing limb angle (TLA) during walking.
• To compare DBO with conventional Bayesian optimization (BO) in a HIL training context.

Main Methods:

• A single-parameter HIL optimization experiment using DBO was conducted.
• Five participants walked on an instrumented treadmill, receiving exoskeleton-applied hip torque pulses.
• The target was to increase the trailing limb angle (TLA); results were compared to a BO control condition.

Main Results:

• Significant increases in TLA were observed post-training in the DBO group, but not the BO group.
• No significant group-level increases in TLA were found during the training phase.
• A weak but significant effect of applied torque on TLA change suggested limited modulation achievability.

Conclusions:

• DBO demonstrated potential for facilitating post-training neuromotor adaptations in TLA.
• Current HIL optimization paradigms may not be sufficient for direct gait training of specific metrics like TLA.
• Future research should explore multi-parameter optimization and alternative propulsion metrics for enhanced gait training.