🌐 English · 한국어
Summary
A Pacific Southwest Region UTC report (PSR-21-16) presenting two studies on how sidewalk autonomous delivery robots (SADRs) affect the safety and comfort of pedestrians and bicyclists on shared-use pathways at Northern Arizona University (NAU), the second U.S. campus to deploy Starship Technologies’ 4-mph six-wheeled robots (March 2019). Study 1 measures objective safety from field video using an adapted post-encroachment time (PET) surrogate measure; Study 2 measures perceived comfort via an original stated-choice survey (PAR-D) administered to a population with three years of real-world SADR experience.
Key Contributions
- First known empirical, real-world safety analysis of SADR interactions with active travelers using a surrogate safety measure.
- A methodological adaptation of PET for vehicles narrower than a travel lane on undelineated shared-use facilities: a static, site-level “bounding box” containing a dynamic, user-level “conflict zone.”
- An original survey instrument (PAR-D) capturing comfort and adoption from people who actually use and encounter SADRs, with stated-choice videos pinned to ~1s (dangerous) and ~3s (moderate) PET.
- Concrete facility-management, route-programming, and robot-design recommendations for safer SADR deployment.
Methodology and Architecture
Study 1 (PET / observed). 10 high-traffic sites (6 north, 4 south campus); HD cameras ~10 ft up; 9am–6pm over 5 days (late Sept/early Oct 2021), reduced to three mealtime/class-transition windows yielding 187 hours of reviewable video. PET = time2 − time1 within a conflict zone (trajectories crossing within ~5s) nested in a per-site bounding box (avg 1,943 sq ft). Two-reviewer consensus; intentional interactions removed. Severity: dangerous (PET ≤ 1.5s), moderate (1.5–3s), normal (>3s). Ordered logit of severity (0/1/2) on conflict- and site-level characteristics (n=201).
Study 2 (PAR-D / reported). Tablet/Qualtrics intercept survey (IRB-approved), 20 questions; sociodemographics, SADR experience/comfort (5-point Likert), and a four-video stated-choice experiment (pedestrian & bicyclist perspectives at ~1s and ~3s PET). Administered at two campus unions over two weeks (April 2022); 522 valid of 526 responses. Four ordered logit comfort models (n=518).
Results
- Study 1: 201 interactions (169 pedestrian, 23 bicyclist, 9 other); mean PET 2.79s; 106 of 192 ped/bike interactions were moderate or dangerous. Pedestrians were in 38/40 (95%) dangerous conflicts, 12 with PET = 0s. Conflicts: 47% crossing, 37% head-on, 15% overtaking; swerving was the most common evasive action. Conflicts clustered at intersecting/narrow paths lacking space delineation. Ordered logit (χ²=28.72, p<0.001): severity ↑ when robot reaches conflict first (β=1.75) and when user 1 swerves (β=1.00); severity ↓ for opposite-direction travel (β=−0.71).
- Study 2: 76% had used Starship delivery; 71% had altered their path because of an SADR. Pedestrians were notably more comfortable than bicyclists. Discomfort sharing paths and having previously altered one’s path predicted lower comfort with evasive maneuvers; higher SADR utilization and stronger future-adoption intent predicted higher comfort; people who bike around campus were less comfortable with the more-severe (1s) bicyclist scenario.
Related Papers
- chen-2018-pedestrian-robot-interaction-experiments-in — complementary pedestrian–robot interaction evidence; Chen’s controlled interaction experiments contrast with Gehrke’s in-the-wild field video, both probing how pedestrians respond to robots crossing or sharing their space.
- rouphail-1998-recommended-procedures-for-chapter-13 — pedestrian capacity/level-of-service framing; relevant to Gehrke’s site-level findings on sidewalk width, intersections, and crowding as drivers of SADR–pedestrian conflict severity.