Go That Way: Exploring Supplementary Physical Movements by a Stationary Robot When Providing Navigation Instructions
Xiang Zhi Tan, Elizabeth J. Carter, Samantha Reig, Aaron Steinfeld · 2019 · Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2019) · doi:10.1145/3308561.3353805
Summary
This paper explores how stationary kiosk-type robots might provide navigation instructions to blind people by supplementing verbal directions with physical movements. The work is inspired by a technique used by Orientation and Mobility (O&M) experts called Palm Drawing, in which the expert traces a route on the person's upturned palm while giving verbal instructions. Using a Rethink Robotics Baxter Research Robot (a human-safe manipulator with 7-degree-of-freedom arms), the researchers developed five methods of delivering physical movement cues: Rotation only (handle rotates to indicate turns), Rotation + Movement (adds translation along the direction of travel), Force (pushes the user's hand in the turn direction), Force + Movement (combines push with translation), and Palm (attempts to replicate the O&M palm-drawing technique by moving a foam-tipped end effector across the user's palm). The robot generated navigation directions from a map server that calculated shortest paths through building hallways, converting route data into verbal instructions like "turn right, walk 20 feet until you reach the water fountain." Two iterative exploratory studies were conducted. In the first (4 blind participants), most found the physical movements distracting and preferred voice-only instructions. Based on this feedback, the second study refined the design by adding a 3D-printed cylindrical handle with a triangular ridge for orientation, switching from logarithmic to linear distance scaling, and adding landmark cues communicated through handle twisting motions.
Key findings
In the second exploration study with 8 blind participants (ages 34-73, all using canes or guide dogs), 5 of 8 preferred the Hand mode (where participants held the robot's handle while it drew the path through arm movements), while 3 preferred Voice-only. No participants preferred the Palm mode. In recall tasks, performance was comparable across modes: Voice participants recalled 4.6 of 5 turns and 3.1 of 4 distances; Hand mode recalled 4.3 turns and 3.4 distances; Palm recalled 4.7 turns and 3.2 distances. The Palm mode failed primarily because the robot's motors had approximately plus/minus 5mm error, making fine palm movements unreliable — participants often could not feel the foam finger's touch or direction. The awkward palm-up position was also uncomfortable. The Hand mode succeeded because it leveraged larger, more perceptible arm movements. Participants valued that the movements helped them form mental maps of routes and that distance differences were communicated through proportional movement lengths. Key concerns included difficulty estimating distances in feet, the need for an introduction to the system before first use, and the challenge of locating the robot in public spaces. Participants were enthusiastic about robot kiosks generally, noting that robots "don't get impatient" and can repeat instructions without annoyance.
Relevance
This research demonstrates both the promise and pitfalls of translating human assistive techniques to robotic platforms. The failure of the Palm mode — a direct recreation of a proven O&M technique — while the less faithful Hand mode succeeded, illustrates that human-robot interactions work best when they adapt human techniques to robot capabilities rather than attempting exact replication. For accessibility practitioners, the finding that physical movement cues can enhance verbal navigation instructions by helping users build mental spatial models has implications beyond robotics — it suggests multimodal direction-giving may be beneficial in any navigation context. The participants' feedback about distance units (feet vs. steps vs. time) echoes a persistent challenge in accessible navigation design, and the observation that landmark-based cues were highly valued reinforces best practices in wayfinding instruction. The practical deployment challenges identified — how do blind users find the robot kiosk, how does the system self-explain without a human introducer — are critical considerations for any public-facing assistive technology. This work contributes to the growing body of evidence that stationary service robots in buildings could serve as accessible information points if designed with disability inclusion from the outset.
Tags: blindness · human-robot interaction · navigation · haptic feedback · orientation and mobility · assistive robotics · indoor navigation