Cognitive Work and Interface Requirements in Autonomous Passenger Mobility: A Taxonomy-Based Analysis of Public Incident Reports
This study analyzes over 2,700 public incident reports to develop a taxonomy-based framework that identifies critical interface requirements for ensuring safety in autonomous passenger mobility by addressing the cognitive work and legibility needs of diverse human actors interacting with automated systems.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
When we imagine a future filled with self-driving cars, the most common picture is one of silence: a vehicle gliding down the road with no one behind the wheel, freeing the human passenger to read, sleep, or simply watch the world go by. It is a vision of automation removing the burden of driving entirely. However, safety experts and engineers increasingly understand that this vision is incomplete. Automation does not simply erase human work; it moves it. Just as a pilot flying a highly automated plane still has a critical job to do when the system encounters a problem, the people around a driverless car—passengers, remote helpers, police officers, and pedestrians—must now perform a new kind of mental labor. They must figure out what the machine is thinking, what it is about to do, and who is in charge when things go wrong. This shift creates a complex web of communication where the vehicle must constantly signal its intentions to a variety of people who are not driving it.
A recent study by Karim Hardy at Embry-Riddle Aeronautical University investigates exactly how this new reality plays out in the real world. Rather than building a new car or running a computer simulation, the researcher looked at thousands of public reports detailing incidents involving autonomous passenger vehicles. These records, which cover everything from robotaxis to autonomous shuttles, serve as a documentary trail of where the interaction between humans and machines has stumbled. By analyzing 2,760 of the latest public reports, the study filtered down to 1,411 specific cases where the problem was not necessarily a mechanical failure, but a breakdown in understanding. The researcher grouped these events into 1,158 distinct clusters to find recurring patterns, treating each report as a clue to the hidden cognitive work required to keep these systems safe.
The analysis revealed that the most frequent trouble spots are not high-speed highway crashes, but rather quiet, low-speed moments where the vehicle is stopped or waiting. Nearly thirty percent of the interface problems occurred when a car was parked, stopped at a curb, or waiting in a lane. In these situations, the vehicle might be doing something perfectly logical, such as waiting for a passenger to load luggage, waiting for a remote operator to give permission to move, or waiting for a pedestrian to cross. Yet, to the people standing nearby, the car's state is often a mystery. Is it broken? Is it yielding? Is it waiting for help? The study found that without clear signals, passengers, pedestrians, and emergency responders are left guessing, and that guessing is where safety risks emerge.
The research also highlighted that the "driver" is no longer a single person in a seat. In many of the reported incidents, the authority to control the vehicle was distributed among several different people. Sometimes a human inside the car was supposed to take over; other times, a remote operator miles away was guiding the vehicle; and in some cases, no human was directly in charge at all, leaving passengers to figure out what to do. The study showed that when these roles are not clearly defined or communicated, confusion spreads. For example, a passenger might not know if they should press an emergency button, or a police officer might not know if it is safe to approach a stopped vehicle. The data suggests that the system fails not because the car cannot drive, but because the people around it cannot read what the car is doing or who is in charge.
From these patterns, the study proposes a new way of thinking about safety. Instead of focusing only on whether the car can navigate a turn or stop for a red light, designers must ensure that the vehicle's state is "legible" to everyone involved. This means the car needs to clearly show its current mode of operation, whether it is waiting, broken, or being helped by a remote human. It needs to tell the passenger exactly what to do if the ride is interrupted. It needs to signal to a police officer whether the vehicle is active or disabled. The researcher argues that safety depends on making these invisible states visible. If a car is in a "degraded" state, meaning it is struggling to function normally, it must communicate that limitation clearly so that people do not expect it to perform like a normal car.
The findings suggest that the future of autonomous mobility relies less on perfecting the driving software and more on perfecting the conversation between the machine and the human world. The study does not claim to have solved these problems, but it provides a clear map of where the misunderstandings happen. By identifying that the most common issues arise from ambiguity in stop-states, unclear authority, and poor communication with emergency responders, the research offers a checklist for improvement. It calls for a design philosophy where the vehicle acts as a transparent partner, constantly updating its human neighbors on its status and intentions. In this view, the success of the self-driving revolution will not be measured by how many miles a car drives without a human, but by how clearly that car can explain itself when it stops.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.