Why Hard Code the Bionic Limbs When They Can Learn From Humans?

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Traditional powered ankle exoskeleton control relies on state machine-based approaches with hard-coded heuristics.
• These traditional methods struggle with the adaptability required for diverse human gait conditions.

Purpose of the Study:

• To propose and validate a task-generic, learning-based model for powered ankle exoskeleton control.
• To enable exoskeletons to adapt to various gait scenarios by learning from human demonstrations.

Main Methods:

• A learning-based model was developed to infer gait constraints from multiple human walking demonstrations.
• The model was validated on ten subjects across different inclines and speeds.
• User studies were conducted with able-bodied subjects performing diverse gait scenarios, including turns and cross-overs.

Main Results:

• The learning-based model demonstrated the ability to generalize to untrained gait conditions, such as higher speeds and inclines.
• Subjects reported comfortable adaptation to various gait scenarios without compromising stability.
• Online experiments confirmed the model's effectiveness across multiple motion conditions.

Conclusions:

• Learning-based control offers a more adaptable and robust solution for powered ankle exoskeletons compared to traditional methods.
• The proposed model successfully supports diverse and untrained gait conditions, improving user experience.
• This approach enhances exoskeleton performance and user stability across a range of walking environments.