Prosthetic Leg Users Start Underconfident, End Overconfident - and Stay Inaccurate Through

Understanding how your own body moves is not as straightforward as it sounds. Athletes spend years calibrating the gap between how a movement feels and what it actually looks like - a gap that coaches, mirrors, and video feedback help close. For someone learning to walk with a robotic prosthetic leg, that calibration process is complicated by something additional: the device attached to the body provides little of the sensory feedback a biological limb would.

A study from North Carolina State University and the University of North Carolina at Chapel Hill, published February 17, 2026 in the open-access journal PNAS Nexus, examined how that perceptual gap evolves during early prosthetic training. The results reveal a predictable pattern - and a perceptual overconfidence that could undermine rehabilitation outcomes if it goes unaddressed.

The Experimental Setup

Nine able-bodied adults participated in a four-day training protocol. To simulate prosthetic ambulation, participants walked with a robotic leg attached to a knee bent at 90 degrees, effectively removing their own lower leg from the locomotion task. They were asked to walk on a treadmill as quickly as possible without touching the handrails - a task that required genuine motor learning over the course of the experiment.

After each practice session, participants viewed computer animations showing biomechanical walking gaits that varied in how natural, balanced, and fluid they appeared. They selected the animation they believed best matched their own recent performance. Their actual gait was measured using standard biomechanical recording. The gap between selected animation and measured performance was the study's core variable.

Day One to Day Four: The Perceptual Arc

On the first day, participants consistently selected gaits that looked worse than their actual performance. They believed they were walking more awkwardly than the sensors confirmed. By day four, the pattern had inverted. Participants selected gaits that looked better than their actual performance.

Actual performance improved substantially across all four days. What did not converge was the relationship between perceived and actual quality. Participants moved from systematic underestimation to systematic overestimation, passing through rough accuracy somewhere in the middle of training.

Helen Huang, the paper's corresponding author and the Jackson Family Distinguished Professor of Biomedical Engineering in the Lampe Joint Department, summarized the arc: "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. The participants were all still inaccurate at assessing the way their own bodies moved - just in a more confident way."

Trunk Over Device: Where Attention Went

Analysis found that trunk lean, step symmetry, and timing dominated participants' self-assessments. What received relatively little attention was the behavior of the prosthetic leg itself - the device participants were explicitly learning to control.

Huang identified the likely mechanism: "They are receiving very little direct feedback about the behavior of the device - they can't see themselves moving." A biological leg sends continuous proprioceptive signals about its position and movement. A robotic prosthetic, with current technology, does not. Without that feedback channel, participants defaulted to interpreting their movement quality through the body parts they could still sense directly.

The Practical Problem with Overconfidence

The shift to overconfidence has direct implications for rehabilitation motivation and effort. A patient who believes their gait is already natural and fluid is less likely to work as hard in subsequent training sessions - even if objective measures show significant remaining room for improvement. In clinical settings where rehabilitation time is limited and costly, that motivational deficit could translate to worse functional outcomes.

The study raises the possibility of real-time perceptual feedback systems - visual displays showing actual gait parameters, haptic feedback from sensors on the device, or augmented reality overlays - that could help users develop more accurate body image alongside motor skill.

Study Constraints

The nine-person sample was drawn entirely from non-disabled adults. Their motivations and psychological state differ substantially from those of actual amputees adjusting to limb loss. The four-day window captures only the earliest phase of what is typically a weeks-to-months rehabilitation process. Whether the same 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 at NC State. The work received support from NIH grant R01HD110519 and NSF grant 2211739.