Locomotion Variability and User Experience in Smart Wheelchair Human–Robot Interaction

Human movement is inherently variable and structured according to task relevance, typically showing higher consistency at task-critical points and greater flexibility in between. However, model-based assistance strategies in human-robot interaction (HRI) traditionally treat human behavior as deterministic, suppressing this natural variability. While this suppression aims to guide users precisely, it often reduces the user's sense of agency and leads to a less favorable interaction experience. This paper investigates how preserving natural movement variability via a novel autonomy-supportive shared control strategy affects both objective task performance and subjective user experience.

The authors implemented a shared control framework on an intelligent powered wheelchair (IPW) where human input controls translational velocity and push forces, while rotational velocity is shared between the user and automation. To study the impact of variability, the system evaluated three conditions: no support (noSup), conventional continuous assistance that suppresses variability (lowVar), and a novel variability-preserving assistance mode (highVar). The highVar mode uses a pure pursuit path-following approach combined with an inverted bell-shaped variability factor function. This design deliberately minimizes robotic rotational correction in task-irrelevant midsections of the path—allowing natural human movement variability—while maximizing guidance at task-relevant start and end points.

A user study was conducted with participants pushing the IPW across 15 repetitions for each of the three conditions. The results confirmed all three study hypotheses. First, the highVar mode successfully generated greater task-irrelevant variability compared to the lowVar mode, validating the control strategy. Second, task-relevant performance remained comparable across both assisted modes, showing that allowing flexibility does not harm precision. Third and most importantly, participants reported significantly higher perceived agency and the highest perceived usefulness in the highVar mode compared to conventional assistance. These findings highlight that assistive robotic systems should respect the embodied structure of human movement rather than treating variability as noise to be eliminated.