Spinal cord injury is devastating, and there is currently no real treatment. Ultimately, we’d want to regrow the spinal cord, but in the meantime, it is now possible to literally reconnect the brain and muscles electronically. I will describe experiments in which we record signals directly from a monkey’s brain, translate them into appropriate control signals, and send them to an electrical stimulator that causes muscles to contract. This “Brain Machine Interface” (BMI) allows voluntary control of paralyzed muscles and could be used to restore movement to human patients with spinal cord injury. Although the technology of existing BMIs, is remarkable, the vast majority require the user to be wired to stationary equipment and allow only intermittent control of a computer cursor or a disembodied robotic limb. The associated control algorithms must be regularly recalibrated to compensate for the changing neurons recorded by the chronically implanted microelectrode arrays. We have developed new methods based on autoencoding neural networks, capable of extracting low-dimensional “latent signals” from wireless neural recordings. The resulting signals appear to be stable over months-long periods. We aim to restore voluntary hand use to the monkeys in their home cages 24 hours a day, without explicit recalibration. The approach should also allow us to study the brain’s representation of movement across a range of motor behaviors that has not previously been possible, and to study the process of adaptation to the prosthesis that a human patient would undergo as we work to move the interface, decoder, and control technology from the lab to the clinic.