Dealing with target uncertainty in a reaching control interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Prosthetic devices require user control via physiological signals.
• Integrating eye movements with other signals enhances prosthetic control.
• Current methods struggle when users view non-target objects or unfoveated targets.

Purpose of the Study:

• To evaluate how target information accuracy affects decoding accuracy with varying neural control signal availability.
• To assess a mixture model that considers both foveated and unfoveated targets.
• To improve prosthetic control for targets not directly fixated.

Main Methods:

• Developed and evaluated a probabilistic mixture model for target selection.
• Varied the accuracy of prior target information.
• Assessed decoding accuracy using natural reaching data and a closed-loop robot-assisted reaching task.
• Incorporated a generic model relying solely on neural signals.

Main Results:

• The mixture model demonstrated effectiveness in scenarios with high target uncertainty.
• Inaccurate target information errors were mitigated by the generic neural-signal-only model.
• Decoding accuracy was influenced by the precision of prior target information.

Conclusions:

• The proposed mixture model robustly handles target uncertainty in prosthetic control.
• Integrating neural signals with a model accounting for potential unfoveated targets improves system performance.
• This approach enhances prosthetic device usability by accommodating diverse user behaviors and visual attention patterns.