| For intracortical brain-computer interface (BCI) controlled neuroprosthetic arms to become a valuable assistive technology for people with upper-limb paralysis they will need to be able to adjust to a number of changes in neural activity that have not previously been well characterized. These include recording instabilities, and changes in neural tuning when interacting with objects. Here I present characterizations of these problems in two human subjects, along with a few solutions.;I quantified the rate at which recorded units become unstable within and between days to inform the design of self-recalibrating decoders. These decoders will provide BCI users with consistent performance, even as units become unstable, by updating to incorporate new units before too many original units have become unstable.;Using the quantification of stability, I also examined whether unit characteristics could predict how long a unit would be stable. I found that units with high firing rates, large peak-topeak voltages, and more accurate tuning were most likely to remain stable. Using this result, future work should be able to create decoders that preferentially rely on stable units in order to enable high-performance BCI control for longer. |
