Cortical control of natural arm movement

摘要

Summary form only given. Our work over the last 20 years has shown that populations of cortical neurons are active during natural arm movements. Furthermore, enough details of these movements can be easily extracted from the single unit spike trains of these neurons to yield high-fidelity, 3D trajectories of reaching and drawing. Over the last nine years, we have developed a technique to chronically implant microwires in the frontal cortices of rhesus monkeys to record 30-90 single units simultaneously for more than three years in the same animal. The recorded signal is processed in real-time and used to control a 3D cursor or robot arm. The animals are able to learn to control these devices while their own arms are restrained and to produce movements that approximate their natural arm movements. These movements are fast, accurate and repeatable over many trials. A key factor in this success is to provide visual feedback to the subject and to have an adaptive decoding algorithm that recognizes learned changes in the animals' recorded neural output. This 'coadaptation' (the animal learns by changing the tuning properties of individual neurons and the extraction algorithm learns by recognizing these changes) shows that high performance can be achieved with small populations of units, easily attained with present electrode technology.