Spinal Cord Stimulation Restores Walking and Sensation in Paralysis Trial

When a spinal cord is completely severed, two things vanish at once. The brain can no longer tell the legs to move, and the legs can no longer tell the brain where they are. That second loss, the disappearance of sensation, is the one that most people outside medicine rarely think about. You cannot walk if you cannot feel the floor.

A clinical trial at Brown University has now demonstrated, for the first time, that both problems can be tackled simultaneously using electrical stimulation. Three volunteers with complete spinal cord injuries walked on a treadmill while implanted electrodes delivered carefully patterned signals above and below their injury sites. Electrodes below the lesion restored enough muscle control for stepping. Electrodes above it produced sensations, felt in the chest, back, or arm, that the participants learned to interpret as information about where their legs were in space. The results, published by Brown University in March 2026, represent a convergence of biomedical engineering, neuroscience, and machine learning that has been decades in the making.

Two arrays, one problem, and a DJ board

The engineering setup is more inventive than it first appears. Surgeons implanted small electrode arrays at two locations on each participant’s spinal cord: one above the injury and one below. The sub-lesion array targeted motor neurons in the lumbar region, the stretch of spinal cord that controls leg movement. By delivering patterned electrical pulses, it activated the neural circuits responsible for flexing and extending the leg, the basic rhythm of a walking stride.

The supra-lesion array, placed above the injury, tackled the sensory problem. Because the damaged cord blocks sensory signals from travelling up to the brain, the researchers used a technique they call sensory replacement. Electrical stimulation above the break produced sensations that participants could feel, such as tingling or pressure in the chest. The intensity of those sensations was linked to the angle of the participant’s knee joint, measured by external sensors. With practice, participants learned to associate a particular chest sensation with a particular leg position. It is not natural proprioception. It is an engineered substitute, and it worked well enough for participants to report accurately where their knees were during walking.

The control interface for the stimulation is a device the research team nicknamed the “DJ board.” It has knobs and sliders that adjust stimulation patterns, intensity, and electrode selection. David Borton, an associate professor of engineering at Brown and the study’s senior author, has spent years refining how electrical parameters map onto biological outcomes. The challenge is scale: each electrode array has multiple contacts, each capable of delivering different pulse widths, amplitudes, and frequencies. The number of possible stimulation combinations is enormous.

That is where machine learning enters the story. An algorithm developed by Thomas Serre’s laboratory at Brown, built by researcher Lakshmi Narasimhan Govindarajan, analysed data from the DJ board sessions and identified which stimulation settings produced the best muscle activation. Rather than testing every combination manually, the algorithm learned to predict effective settings, dramatically reducing the time needed to calibrate each participant’s system.

The team plans to recruit new participants for a longer-term study that will test stimulation outside the hospital, in homes and rehabilitation centres. Whether this technology can scale from a three-person trial to a clinical product depends on manufacturing the electrode arrays reliably, making the machine-learning calibration fast enough for routine clinical use, and securing funding for the regulatory pathway. The gap between a treadmill in a laboratory and independent mobility in the real world remains wide. But for three people who had been told they would never feel the ground again, that gap just narrowed.