How Should Robotaxis Behave Before Emergency Cones Arrive?
Before traffic cones appear, an emergency scene is mostly clues: smoke, flashing lights, people moving against normal traffic, a firefighter’s hand. NHTSA says a Zoox robotaxi entered a smoke-obscured fire scene on June 20, braked hard and stopped. A remote tactician guided it backward. Only then did responders place cones across two lanes. That order matters. Cones are the punctuation at the end of the warning. The dangerous minutes come before the street has been neatly marked. ‘Better smoke detection’ is not a complete public answer. A robotaxi needs an emergency-scene posture people can predict: slow early, yield to responders and bring in remote help without waiting to name the scene perfectly. But holding still in smoke can be dangerous too. Before the cones arrive, what should take priority: clearing the lane, holding position or following the first responder who is in view?
Comments
The first responder in view should take priority—but only if every robotaxi understands the same few commands. A firefighter in smoke should not have to guess whether a raised palm means stop to Zoox but nothing to another car. Stop, reverse and clear the lane need shared signals, and the car should show which command it understood. Otherwise the person already managing the emergency gets one more machine to manage.
The car’s answer has to be outside the car. A firefighter should get one unmistakable state: STOPPING, HOLDING or REVERSING, with the direction clear before the wheels move. If they have to stare at the wheels to work out whether the raised palm registered, the interface has already failed.
The filing supports one behavior failure, but it leaves Mina’s shared-signal idea open. It says the empty vehicle failed to detect and respond to heavy smoke, entered the scene, braked hard, then reversed under remote guidance. It does not say a firefighter signaled the car, or that the July 15 update can read hand gestures. The remedy description is thin too: better smoke detection plus measures to avoid active fire scenes, with no public scenario results. So yes, test common responder commands—but do not let that replace earlier detection and a predictable fallback. Publish results for smoke before cones, obscured flashing lights, a responder in the lane and lost remote contact. A firefighter should not discover which cue works while standing in traffic.
Remote help is useful. It can also make a weak fallback look healthier than it is. Publish the time from the hard brake to remote guidance, then test what the car does when that link is slow or dead. Fire crews should not have to discover whether “remote tactician” means three seconds or three minutes while standing in smoke.