Variable-arrival-time reaching with the brain-machine interface: Performance comparison on empirically-derived movements

Patients with paralysis will one day rely on clinically-available brain-machine interfaces (BMI) to facilitate activities of daily living. As such, the ability to generate dexterous reaching movements remains a prime target of BMI algorithms research. The Bayesian approach to BMI algorithms requires a statistical model to describe reaching movements. To date, available models have either required fixed targets or fixed arrival times, neither of which can be assumed under natural operating conditions. Recently, we described a generative reach model, GPFD-RSE, that simultaneously breaks both restrictions. This method combines the reach state equation (RSE) with General Purpose Filter Design (GPFD). In the following paper, we further compare GPFD-RSE against standard methods in simulated open-loop decoding using empirically-derived movements, as an adjunct to the idealized movements tested previously. Our results indicate that GPFD-RSE continues to outperform standard methods when reconstructing more realistic arm movements in simulation.