Make autonomous driving predictable in rare road events

Handle long-tail situations such as construction detours, occluded pedestrians, unusual vehicles, and negotiations among road users, so autonomous systems can slow safely or request takeover when uncertain.

Autonomous drivingLong-tail scenariosRoad safety

Current progress and boundaries

Autonomous vehicles are tested and operated in limited areas and conditions, but open-road scenarios are much more complex than routine routes.

Why it remains difficult

Rare events cannot be exhaustively enumerated, and perception errors affect planning and control. Road rules, vulnerable road users? behavior, and assignment of responsibility must also be considered.

Paths worth exploring

Build a reproducible library of long-tail scenarios and combine simulation, closed-course, and controlled-road testing. Define minimum-risk states and boundaries for remote assistance.

How to evaluate success

Track serious conflict rates, takeover quality, scenario coverage, and safety-boundary violations, with independent review of crashes and near misses.