Learning by Demonstration for Motion Planning of Upper-Limb Exoskeletons

Area of Science:

• Robotics
• Human-Robot Interaction
• Rehabilitation Engineering

Background:

• Traditional motion planning for upper-limb exoskeletons often fails to ensure anthropomorphic criteria, especially with non-redundant systems.
• Existing methods struggle to adapt to unstructured environments and patient-specific needs during Activities of Daily Living (ADLs).

Purpose of the Study:

• To propose and validate a Learning by Demonstration-based motion planning system for upper-limb exoskeletons.
• To ensure anthropomorphic criteria are met across the human-robot workspace during ADLs.
• To enhance exoskeleton assistance for patients in unstructured environments.

Main Methods:

• The system combines Learning by Demonstration with Dynamic Motion Primitives and machine learning.
• Task- and patient-specific joint trajectories are generated based on learned movements.
• Validation involved simulations and real-world testing with upper-limb and wrist-hand exoskeletons and patients with Limb Girdle Muscular Dystrophy.

Main Results:

• A 100% success rate was achieved in task fulfillment for chosen ADLs (drinking, pouring, lifting).
• The system demonstrated high generalization capabilities concerning environmental variability.
• Exoskeleton configurations consistently adhered to anthropomorphic criteria.

Conclusions:

• The proposed motion planning system effectively assists patients with upper-limb exoskeletons during ADLs in unstructured settings.
• The integration of Learning by Demonstration ensures anthropomorphic movements and adaptability.
• This approach offers a significant advancement in personalized and natural human-robot interaction for rehabilitation.