Designing closed-loop brain-machine interfaces with network of spiking neurons using MPC strategy

Abstract

Brain-machine interfaces (BMIs) are human-machine integration systems that provide an interface between the brain and a machine to sense cortical neuronal activity for the purpose of restoring impaired motor tasks. In our previous work [1], an optimal design of BMIs based on artificial sensory feedback was developed using model predictive control which relied on neuronal activity in the form of spiking. From a real implementation perspective, a more generalized framework that utilizes spiking is proposed in this paper. Specifically, a charge-balanced intra-cortical micro-stimulation (ICMS) current and a network of spiking neurons are adopted to compensate the lost feedback information. Next, an artificial sensory feedback framework using the network of spiking neurons is designed based on model predictive control (MPC) strategy, and an optimization problem is formulated according to this framework. Since the charge-balanced ICMS current is composed of several integer parameters, the optimization problem also includes some integer decision variables and is hard to be solved. In this paper, a heuristic population-based search algorithm called particle swarm optimization (PSO) algorithm is used to solve this optimization problem. Considering the updated particles may violate the input constraints, additional constraints are designed to guarantee that the decision variables can satisfy the input constraints. Finally, simulation results show the effectiveness of the designed closed-loop BMIs during recovery of natural performance.