Rider trust and the role of the operator in automated shuttles
archive: archived pipeline: cataloged verified
Summary
Survey and observational study of rider trust in an EasyMile Level-4 autonomous shuttle deployed at the University of Utah as a first-mile/last-mile pilot. Pre- and post-ride Likert surveys (n=70 pre, n=236 post; 306 total respondents) captured expectations, experience, and trust-related attitudes; GoPro video of 19 repeat riders was coded in BORIS for trust-indicative behaviors (casual conversation, anticipatory statements, cell-phone use, informational questions). Repeated ridership predicted more positive experience (tau_b=.164), better expectation alignment (tau_b=.252), and greater destination confidence (tau_b=.216). Experiencing one or more emergency stops (e-stops) was associated with more negative experience, lower display-cue confidence, and perceived need for operator intervention. Video coding showed repeat riders engaged in more casual conversation and made more anticipatory statements about shuttle behavior. The onboard operator spent the largest portion of time in conversation with riders and actively worked to foster trust, raising questions for fully driverless deployment.
Key finding
Rider trust in low-speed autonomous shuttles grows with repeated ridership but is undermined by unexpected e-stops; the human backup operator plays a central, trust-building role that must be replaced by other channels (displays, signage, auditory cues) before fully driverless operation is viable.
Methodology
Field study at University of Utah pilot route (~0.5 mile loop, 3 stops, ~15 min/loop) over 4 weeks. Pre- and post-ride iPad surveys with 7-point Likert items in mirrored positive/negative variants to detect inattentive responding. GoPro video of repeat riders coded in BORIS for state and point behaviors. Survey associations analyzed with Kendall's tau_b due to heteroskedasticity and ties; pre/post comparisons via two-sample t-tests. E-stop occurrences obtained from EasyMile telemetry.
Sample size: 306 survey respondents (236 post-ride, 70 pre-ride); 19 repeat riders coded from video
Quality score: 4 / 5