Preserved Neural Dynamics across Arm- and Brain-controlled Movements

Abstract

The neural activity for motor control is complex and dynamic; it has been found to dramatically transit from planning to executing movements. As brain-machine interfaces (BMIs) can directly connect the brain and the external world by yielding comparable motor outcomes with artificial apparatus, a central question is whether the BMI-controlled movements share neural dynamics or underlying mechanism with natural movements. To enable a systematic comparison, we developed a feedforward BMI framework with distinct planning and executing epochs that enables ballistic cursor control to intercept moving targets. This BMI allowed monkeys to voluntarily initiate neural states which controlled the direction and timing to launch a ballistic movement, like skeet shooting. Based on this, we found similar neural representations and computational structures across arm- and brain-controlled conditions. Notably, in addition to resembling the rotational structure in natural reaching, the neural population dynamics during open-loop BMI also shared preserved manifolds with those during reaching arm movements. These findings suggest a fundamental principle, and reveal a set of basic computational motifs for the neural control of movement in an abstract hierarchy in the absence of constraints from actuators. This study thus has the potential to reshape the consideration of how BMIs assist paralyzed patients in interacting with dynamic environments, and to promote next-generation BMI systems.