CaBot: Designing and Evaluating an Autonomous Navigation Robot for Blind People
João Guerreiro, Daisuke Sato, Saki Asakawa, Huixu Dong, Kris M. Kitani, Chieko Asakawa · 2019 · Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2019) · doi:10.1145/3308561.3353771
Summary
This paper presents CaBot (Carry-on roBot), an autonomous suitcase-shaped navigation robot designed to guide blind people to destinations in unfamiliar indoor environments while avoiding obstacles. Unlike smartphone navigation apps that provide turn-by-turn instructions but leave obstacle avoidance and path execution to the user, and unlike guide dogs that handle local navigation but depend on the user to know the route, CaBot combines both capabilities: it plans routes using a pre-mapped floorplan, localises itself using LiDAR and wheel odometry via ROS (Robot Operating System), detects and avoids dynamic obstacles using a stereo camera with YOLOv2 object recognition, and conveys directional information through vibro-tactile feedback on a modified suitcase handle. The suitcase form factor was chosen to blend into public environments without attracting unwanted attention, and to enable side-by-side walking (CaBot on the user’s left, slightly ahead) mimicking the guide dog or sighted guide experience. The handle contains three vibrators (left, right, top) driven by an Arduino Nano, producing distinct patterns: four consecutive vibrations for regular turns, two for slight turns (obstacle avoidance), and a single longer vibration on top for moving forward. Speed is user-controlled via two buttons on the handle (increments of 0.05 m/s, maximum 1.0 m/s), with automatic deceleration before turns and stops when paths are blocked. Speech feedback via bone conduction headphones conveys environmental semantics — points of interest, landmarks, floor changes, and explanations for deviations (e.g., "avoiding a person"). A key technical contribution is the robot-human centred footprint adaptation, which shifts the path planner’s control centre to account for the combined space of robot and human, enabling navigation through narrower passages without overestimating the pair’s footprint.
Key findings
A user study with ten blind participants (ages 38-73, five guide dog users and five white cane users) evaluated vibro-tactile perception, speed preferences, the impact of vibro-tactile feedback on navigation confidence, and overall experience across routes of 43m, 41m, and 89m (round trip) with turns and human obstacles. All participants identified vibration patterns with 100% accuracy both standing and walking. Preferred walking speeds ranged from 0.70-1.00 m/s, with all guide dog users selecting the maximum (1.00 m/s). CaBot completed all navigation tasks with zero navigation errors (no wrong turns, collisions, or missed destinations). The SUS usability score averaged 88 (SD=20.71) — well above the "excellent" threshold — with 9 of 10 participants scoring at least 90. Users’ confidence (6.7/7), safety (6.3/7), and trust (6.4/7) with CaBot were comparable to their confidence and safety with their own navigation aids in familiar locations, and significantly higher than with their aids in unfamiliar locations (p<0.05 for confidence). Critically, all three measures significantly exceeded participants’ pre-experiment expectations of what a navigation robot would be like (p<0.01). Participants frequently compared CaBot to a guide dog or sighted guide: P3 noted the handle’s haptic feedback felt like "the strength of a dog’s pull and the pressure that you feel on your hand," while P9 declared "I’ll be calling my dog CaBot later!" Vibro-tactile feedback did not significantly improve navigation outcomes (since CaBot is autonomous, feedback doesn’t change performance) but was valued by six participants for anticipating the robot’s actions.
Relevance
CaBot represents the most complete autonomous navigation robot evaluation with blind users published to date, combining autonomous path planning, real-time obstacle avoidance, environmental semantics, haptic directional feedback, and user-controlled speed in a socially acceptable form factor. The finding that CaBot’s performance "highly exceeded users’ expectations" and was compared to guide dogs and sighted guides suggests that autonomous navigation robots are approaching the quality threshold needed for real-world adoption. For accessibility practitioners, the design decisions are instructive: the suitcase form factor for social acceptability, left-side positioning following guide dog convention, bone conduction audio to preserve environmental awareness, vibro-tactile rather than audio directional cues (works in noisy environments and reduces cognitive load), and the robot-human centred footprint for efficient path planning. Participants envisioned deployment in airports, shopping malls, hospitals, and universities — large unfamiliar indoor spaces where both white canes and guide dogs are limited. The key remaining challenges are reducing size and weight (25.2 kg), extending battery life, and handling dynamic outdoor environments.
Tags: navigation robot · blind · orientation and mobility · haptic feedback · LiDAR · computer vision · human-robot interaction · obstacle avoidance · bone conduction · indoor navigation