Grasp Prediction Toward Naturalistic Exoskeleton Glove Control

Area of Science:

• Robotics
• Human-Computer Interaction
• Biomechanics

Background:

• Exoskeleton systems primarily focus on mechanical design and actuation for rehabilitation.
• Intelligent grasp assistance through user intent estimation remains an under-addressed challenge in exoskeleton control.
• Mapping complex hand motions to a latent space can simplify prediction by reducing noise and variables.

Purpose of the Study:

• To present accurate grasp prediction algorithms for naturalistic, synergistic control of exoskeleton gloves.
• To address the gap in estimating user intent for intelligent grasp assistance.
• To develop algorithms that minimize user input for effective exoskeleton glove operation.

Main Methods:

• Developed two latent space grasp prediction algorithms for exoskeleton glove control.
• Algorithm 1: Linear regression to determine slope and prediction horizon.
• Algorithm 2: Gaussian process trajectory matching for probabilistic motion prediction.

Main Results:

• Both algorithms were tested on published motion data from healthy subjects.
• Experimental validation using the RML glove showed prediction accuracy comparable to simulation results.
• The algorithms demonstrated effectiveness in predicting grasp intentions for exoskeleton control.

Conclusions:

• The proposed grasp prediction algorithms can serve as a foundation for advanced exoskeleton control systems.
• These algorithms facilitate synergistic control, amplifying user motion while guiding desired grasps.
• Future research directions include integrating these algorithms into shifting authority controllers for enhanced assistive capabilities.