How the Brain Adjusts to a Robotic Leg - and Why It Gets Overconfident

Learning to walk with a robotic prosthetic leg involves two parallel processes that do not always stay in sync. One is motor learning - the gradual improvement in gait mechanics that comes from practice. The other is perceptual calibration - developing an accurate sense of how your body is actually moving. A study from North Carolina State University suggests these two processes diverge in a predictable and potentially problematic way.

The study, published February 17, 2026 in PNAS Nexus, recruited nine non-disabled adults with no prior experience using any prosthetic device. Over four days, participants walked on a treadmill using a robotic prosthetic leg attached to a knee bent at a 90-degree angle - their own leg effectively removed from the locomotion task. After each practice session, they were shown computer animations displaying a range of biomechanical walking gaits and asked to identify which most closely matched what they had just been doing.

From Underestimation to Overconfidence

At the start of training, participants consistently underestimated their performance. They selected gaits that looked more awkward and off-balance than their actual biomechanical recordings showed. By the end of day four, the pattern had reversed: participants selected gaits that looked more fluid and natural than their real performance justified.

Actual gait performance did improve significantly across the four days. But participants' internal sense of their own movement quality remained systematically inaccurate throughout. They moved from one type of error to the opposite type, passing through rough accuracy briefly in the middle of training.

Helen Huang, the study's corresponding author and the Jackson Family Distinguished Professor of Biomedical Engineering in the Lampe Joint Department at NC State and the University of North Carolina at Chapel Hill, described the trajectory: "Initially, participants felt their gait was more off-balance and stilted than it actually was. By the end of the four-day study, participants felt their gait was more fluid and natural than it actually was - just in a more confident way."

What Participants Actually Focused On

Analysis found that four gait features dominated participants' self-assessments: trunk lean, step length of the non-prosthetic leg, prosthetic leg step time, and step time symmetry index. Participants placed relatively little weight on the behavior of the prosthetic device itself - the very component they were learning to control.

Huang identified a likely reason: "They are receiving very little direct feedback about the behavior of the device - they can't see themselves moving." Without sensory feedback from the prosthetic leg comparable to the proprioceptive signals a biological limb provides, participants defaulted to interpreting their movement quality through the parts of their body they could still feel directly.

Why Overconfidence Matters for Rehabilitation

The overconfidence finding at the end of training carries specific clinical implications. A patient who believes their gait is already natural and fluid has reduced motivation to continue working on improvement, even when objective measurements show room for significant gains. In a rehabilitation setting with limited session time and resources, that perceptual error could translate directly into worse long-term outcomes.

The study raises the possibility of addressing this through real-time feedback systems - visual displays, haptic signals, or auditory cues - that give prosthetic users accurate information about their actual movement patterns during training. Current prosthetic devices provide limited sensory feedback compared to biological limbs, and this study suggests that gap affects not just motor learning but the development of accurate body image.

Study Constraints

The sample of nine participants is small, and all were non-disabled adults in an experimental context. Their experience differs in important ways from that of amputees who face distinct psychological challenges around body image following limb loss. The four-day training period captures only the early phase of prosthetic learning. How body image accuracy evolves over weeks or months of continued use remains to be studied. Whether the same underestimation-to-overconfidence arc appears in clinical populations requires separate investigation.

First author I-Chieh Lee was a research associate professor in the Lampe Joint Department at the time of the study. Co-authors include Huan Min and Ming Liu, both from NC State. The work was supported by NIH grant R01HD110519 and NSF grant 2211739.