Parallel command principle in sensorimotor control for brain-inspired AI design

The cortex to brainstem connection is a major conduit for sensory and motor control in the mammalian neocortex, yet its connectional architecture and organizational principles remain incompletely characterized. In this study, we reconstructed the single-cell projectome of cortical neurons in the mouse brain and systematically mapped the axonal projection patterns of cortical pyramidal tract neurons in the brainstem. Using this approach, we identified a parallel projection architecture in which distinct neuronal clusters independently targeted specific brainstem nuclei with limited projection overlap. Notably, neurons in the specialized anterior cingulate area exhibited segment-specific projections to the reticular formation, supporting the balance between specific and integrative motor control. In addition, we observed two segregated modules in the midbrain that separately received somatosensory and exteroceptive sensory inputs from the mouse cortex. Translating these connectivity principles into artificial neural networks, we designed a novel feedback architecture that lowers the training loss by 21% compared to conventional feedforward models and improved robustness during error recovery. Overall, our findings elucidate the mesoscale connectivity of the cortical pathway and provide a valuable reference for optimizing brain-inspired artificial intelligence using connectomic design principles.