A robotaxi can perform better than a human driver in average conditions and still create serious risks when normal traffic rules no longer provide a clear answer. That is why future automated-vehicle standards should address exceptional situations as well as routine driving.
Streetsblog reported this summer that Waymo robotaxis were involved in 68 percent fewer police-reportable crashes than human drivers, citing an independent study by the Insurance Institute for Highway Safety. However, crash statistics do not necessarily capture street-level problems such as vehicles blocking bike lanes, obstructing emergency responders, passing school buses or creating other hazards.
That limitation has gained attention from regulators. In July, the National Highway Traffic Safety Administration said it had identified a pattern of driverless vehicles interfering with law enforcement and other first responders. Later that month, the agency announced that it was accelerating work on the first federal performance standards for automated vehicles.
Those standards should establish what a robotaxi is permitted to do when road conditions depart from ordinary expectations. An ambulance may need a vehicle to clear an intersection, while a police officer or construction worker may direct traffic in a way that conflicts with a signal. Cyclists and pedestrians may also have to navigate around a vehicle stopped in an inappropriate location.
These cases require more than basic driving ability. Vehicles need defined rules for when to continue autonomously, when to slow or stop, and how an authorized person can override normal behavior. Regulators do not necessarily need access to every line of software, but they should receive evidence that a vehicle can identify situations outside its normal operating scope, choose a safe default response and allow reliable intervention.
Manufacturers should identify high-risk conditions in which standard traffic rules may conflict with emergency instructions, temporary traffic controls or the safety of people outside the vehicle. They should then demonstrate how the vehicle responds and what happens when its confidence falls.
Remote assistance should also have clear boundaries. Operators helping a stuck vehicle should have specifically defined powers, with records showing which actions they can authorize and a mechanism for withdrawing that authority if necessary.
The proposal is consistent with a principle highlighted in a peer-reviewed analysis of explainable artificial intelligence by Plamen Angelov and colleagues: systems should recognize situations they were not designed to handle reliably. Applied to transportation, that means a robotaxi should detect when a situation has moved beyond ordinary driving and follow a tested procedure.
Cities, regulators and road users do not need a technical manual, but they do need a plain-language explanation of how robotaxis behave around first responders, temporary traffic controls, blocked lanes, school buses, cyclists and pedestrians. Future AV standards should therefore evaluate both everyday driving performance and the vehicle’s ability to respond safely when ordinary rules are no longer enough.
Sources: Streetsblog USA.