Certifying the people who fly the robotaxis

The vehicle has a safety case. Does the operations center?

Usually not a comparable one. Enormous rigor goes into the vehicle's safety argument — hazard analysis, redundancy, validation mileage, all of it defensible. Then the vehicle encounters something outside its operational design domain and asks a human for guidance, and that human is working a shift pattern nobody analyzed, on an interface nobody hazard-assessed, under a queue depth nobody bounded.

The remote operator is part of the safety-critical system. If we are honest about that, their competence, their workload and their authority all belong inside the safety case.

“The failure mode is not panic. It is agreeing with the system too readily at the end of a long shift.”

What would a competence standard cover?

Situational assessment from degraded sensor views. Knowing the limits of your own understanding of the vehicle's state estimate. Escalation discipline: when to stop the vehicle rather than resolve the situation cleverly. And explicit training in automation bias, because the failure mode is not panic — it is agreeing with the system too readily at the end of a long shift.

Aviation solved a version of this decades ago with crew resource management. We are re-learning it, slowly, with worse instrumentation of human factors than aviation had.

• Assessed proficiency in degraded-information decision making.
• Bounded queue depth and workload limits written into the safety case.
• Recurrent training with scenario assessment, not annual slide decks.
• Explicit authority to stop a vehicle without commercial escalation.

Where should ACS focus?

Three things. Publish a competence standard for remote operations that fleets can adopt voluntarily before anyone mandates one badly. Train assessors who can evaluate a safety case that includes human elements, which is a genuinely different skill from evaluating software. And give operators a professional identity, so that a person who refuses to release a vehicle is exercising a recognized duty rather than risking their job.

The Autonomy and Robotics specialist group has started the first of these, and I would like it finished before an incident writes the standard for us.