Synthetic Data-Driven Exoskeleton Control via Contralateral Gait Fusion for Variable-Speed Walking.

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Data-driven exoskeletons promise enhanced human mobility but face challenges in data collection and tuning.
• Current methods require extensive manual effort, limiting widespread adoption.

Purpose of the Study:

• To develop an efficient synthetic data generation pipeline for data-driven exoskeleton control.
• To enable real-time, adaptive torque assistance for variable-speed walking.

Main Methods:

• Leveraged an Adversarial Motion Priors (AMP) agent in physics-based simulation to generate high-fidelity walking data.
• Developed a CNN-Transformer architecture for end-to-end torque prediction from contralateral swing-phase data.
• Validated the approach using a custom ankle exoskeleton for sim-to-real transferability.

Main Results:

• Achieved low prediction error (RMSE ~0.081 Nm/kg, R² ~0.836) across various walking speeds (0.6-1.75 m/s).
• Demonstrated significant reduction in user ankle mechanical work (up to 14%).
• Showcased inherent fault tolerance in the multi-sensor configuration for safe operation.

Conclusions:

• The synthetic data approach provides a scalable pathway for autonomous exoskeleton deployment.
• This method facilitates practical, real-world application of adaptive exoskeleton assistance.
• The system offers robust performance and enhanced safety features.