Autonomous Prosthetic Arms Feel Most Like Part of the Body When They Move at Human Speed
When a prosthetic arm moves on its own, without the user consciously directing it, how it moves matters as much as whether it moves accurately. The question is not only functional - does it reach the target? - but psychological: does the person wearing it experience it as part of their body, or as something foreign that happens to be attached to them?
As AI-powered prostheses capable of autonomous or semi-autonomous movement become increasingly realistic, researchers need to understand what movement parameters make a robotic limb feel like an extension of the self rather than a well-engineered external tool. A study published in Scientific Reports examined one of the most basic design parameters: how fast the arm moves.
The Experimental Design
Harin Manujaya Hapuarachchi - then a doctoral student, now an assistant professor at Kochi University of Technology's School of Informatics - and colleagues at the Toyohashi University of Technology used virtual reality to simulate a situation in which a participant's own arm was replaced by a robotic prosthetic. Participants performed a reaching task in which the prosthetic arm flexed autonomously toward a target. Movement duration was systematically varied across six levels, ranging from 125 milliseconds to 4 seconds, spanning well below natural human reaching speed to well above it.
After each condition, participants rated four dimensions: body ownership (does this feel like my arm?), sense of agency (do I feel in control of the movement?), usability using the System Usability Scale, and social impressions of the robot using the Robotic Social Attributes Scale, which measures perceived competence, warmth, and discomfort.
What the Results Showed
The findings were consistent and clear. At a movement duration of approximately one second - close to the natural duration of a human reaching movement - body ownership, sense of agency, and usability scores were all at their highest. Both extremes performed significantly worse. At 125 milliseconds, the arm moved faster than any natural human reach. At 4 seconds, it moved at roughly one-quarter natural speed. In both cases, body ownership, agency sense, and usability fell substantially.
The pattern for social perception followed a related but distinct curve. Perceived competence was highest at moderate to slightly faster speeds - users perceived a somewhat fast robot as more capable. But discomfort peaked sharply at the fastest condition, 125 milliseconds, where the unexpected speed of an autonomous movement created what participants experienced as unsettling. Perceived warmth showed no clear dependence on movement speed.
Why Speed Matters for Embodiment
Prior work had established that when a limb moves on its own, people can come to experience it as less foreign when the movement's intention is legible - when they understand what the arm is trying to do. This study extends that finding: even when intention is clear, speed shapes whether the movement reads as natural or uncanny. A movement that is too fast to match the brain's predictions about bodily motion breaks the sense that the arm belongs to the body, regardless of whether it reached the correct target.
This has design implications that extend beyond standard prosthetics. Supernumerary robotic limbs - additional robotic arms attached to the body for augmentation rather than replacement - exoskeletons, and wearable robots all involve movement that the user may not be directly controlling. The finding that moderate, human-scale speed maximizes acceptance and usability suggests that performance optimization should account for embodiment, not just accuracy and speed.
Limitations and Next Steps
The study used virtual reality rather than physical prosthetics, which means participants were not experiencing the weight, thermal properties, or mechanical feedback of an actual device. VR enables safe testing of designs and control schemes before they exist in physical form, but the psychological response to a virtual prosthetic and a physical one may differ. The sample consisted of able-bodied individuals rather than prosthesis users, whose experience of body ownership may be shaped differently by prior device use.
The research team plans to examine how extended daily use affects these preferences. Tools that initially feel unnaturally fast may become normal over time as users adapt - a question relevant to whether the optimal speed for initial adoption is also optimal for long-term experience. The study was funded by JSPS KAKENHI (JP22KK0158), the Murata Science and Education Foundation, JST (JPMJFS121), and MEXT (202334Z302).