Motor cortical decoding performance depends on controlled system order.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Control Theory

Background:

• Intracortical brain-machine interfaces (BMIs) decode intended movements from neural activity.
• Kalman filters are commonly used state estimators for BMIs, assuming identified models capture closed-loop dynamics.

Purpose of the Study:

• To re-examine assumptions of Kalman filters in BMIs.
• To investigate how changing task dynamics affect neural coding and BMI performance.
• To propose principles for enhancing BMI decoder robustness.

Main Methods:

• Partitioning closed-loop models to separate brain feedback policies from interface and task dynamics.
• Conducting experiments where system order (position vs. velocity control) is manipulated in manual tasks.
• Analyzing neural coding changes in response to altered system dynamics.

Main Results:

• Closed-loop models can be decomposed, revealing distinct brain control policies.
• Altering system dynamics (e.g., from position to velocity control) changes neural coding.
• Performance degradation in BMI decoders is observed when transitioning from manual to BMI control with different dynamics.

Conclusions:

• Brain control policies adapt to task dynamics.
• Mismatch in system order between manual and BMI control can impair decoder performance.
• Improving BMI performance requires identifying brain controller dynamics and matching system orders.