Brain-computer interfaces (BCIs) aim to help people with paralysis to improve their communication and independence. Intracortical BCIs (iBCIs) have shown promising results in pilot clinical trials. Despite the performance improvements over the last decades, BCI systems still make errors that need to be corrected manually by the user.
While much work continues to be done to increase iBCI performance by improving the accuracy and reliability of movement intention decoders, here we explore a complementary and less explored approach: have the iBCI automatically identify, using neural activity, when an error occurs. This allows the system to then automatically undo the error. This ‘automatic error detect-and-undo’ strategy takes advantage of the closed-loop nature of a BCI; the user has constant visual feedback and is aware of when the BCI performs an unintended action. Somewhere in the brain, the user’s neural activity will reflect this detection and recognition of an error. In our recent work with monkeys (Even-Chen et al. 2017), we have detected a putative error signal in motor cortex and showed its utility in real-time iBCI. Here, we investigated the feasibility of a similar detect-and-undo system for iBCI using intracortical recordings from two human participants with Utah arrays implanted in motor cortex.
We found that motor cortex neural activity is modulated by error occurrence. Using spikes and local field potentials we were able to detect errors with high accuracy (70–85%) with minimal misdetections (0-3%). Last, we discuss the design parameters of an error detect-and-act system, including the appropriate corrective actions and the right time for an intervention. Using simulations, we estimated the appropriate intervention time and predicted that in difficult task performance can be improved two-fold.
These results suggest that detecting and undoing errors in real-time can make hard tasks feel easier, increase iBCI performance, and improve the user experience.