A team led by UC Irvine, with Caltech and the Keck School of Medicine of USC, has demonstrated a bidirectional brain-computer interface that simultaneously decodes walking intent from the motor cortex and delivers artificial leg sensation through electrical stimulation of the sensory cortex. A 50-year-old woman operated an FDA-approved robotic exoskeleton using the system, completing 10 exercises with no adverse events. The results, published in Brain Stimulation, represent the first time both motor control and sensory feedback have been achieved in a single BCI system for walking.
Robotic exoskeletons already exist for gait rehabilitation, but current devices rely on manual control — joysticks, buttons, or therapist input — and provide no sensory feedback to the user. The person wearing the exoskeleton cannot feel their legs or the ground. An Do, an associate professor of neurology at UC Irvine who led the study, says the absence of sensation is a significant limitation. Without it, the user has no proprioceptive sense of where their legs are or what they are doing, which limits confidence, safety and the naturalness of movement.
The system uses bilateral interhemispheric electrocorticography arrays — electrode grids placed along the midline fissure between the two hemispheres, directly over the leg areas of the motor and sensory cortices. This placement is itself a technical contribution. Leg representation in the brain sits deep in the interhemispheric fissure, an area that conventional lateral electrode placements cover poorly. Charles Liu, professor of neurological surgery at USC and director of the USC Neurorestoration Center, says the interhemispheric approach provides more robust and reliable neural signals associated with leg movements than lateral alternatives.
On the motor side, the system decodes stepping intent from the electrocorticography signals in real time, achieving a correlation of 0.92 between decoded and actual motor signals. On the sensory side, it delivers targeted electrical stimulation to the somatosensory cortex to produce the sensation of stepping in each leg. In a blind step-counting task, the participant identified steps with 93 percent accuracy. She discriminated right-leg, left-leg and null sensations with 96, 84 and 100 percent accuracy respectively.
The entire system ran on three 48-megahertz microcontrollers embedded in a portable unit, with no tethered external computer. Jeffrey Lim, a postdoctoral scholar in biomedical engineering at UC Irvine and the paper’s first author, says portability is a prerequisite for any system that is going to be practical outside a laboratory. The exoskeleton used was the Ekso GT, an FDA-approved powered device made by Ekso Bionics.
The participant was not paralysed. She was a 50-year-old woman undergoing epilepsy evaluation who already had bilateral subdural electrode grids implanted for clinical monitoring — a standard procedure in epilepsy care that gave the team temporary access to the cortical surface. This is a common and pragmatic route for early BCI locomotion studies: patients who need intracranial electrodes for epilepsy mapping offer a brief window in which researchers can test motor and sensory interfaces without additional surgical risk.
The study involved a single participant over a limited period, and the researchers are explicit about what comes next. Payam Heydari, a professor of electrical engineering at UC Irvine, says the team is working toward a fully implantable version with skull-mounted and chest-wall components that would eliminate the transdermal connections that carry infection risk in current setups. Zoran Nenadic, professor of biomedical engineering at UC Irvine, says the ultimate goal is to test the system on people with complete leg paralysis to confirm that the sensorimotor loop can function when the spinal cord pathway is fully interrupted.
The work was funded by the National Science Foundation. Collaborators included Richard Andersen’s laboratory at Caltech and clinicians at Rancho Los Amigos National Rehabilitation Center. The paper lists 18 authors across the four institutions.