Beyond Default Behaviors: Reimagining Autonomous Mobility for Safe Interaction

Autonomous vehicles are engineered around a default user, a default mechanism for directives, and a default fallback: an able-bodied pedestrian who inputs a general dropoff location, and a full stop as the answer to uncertainty. This extended abstract draws a line through three AV research threads that each begin by refusing one of those defaults: (i) sidewalk-accessibility perception learned directly from mobility aid users, using per-annotator disagreement as training signal rather than a majority-collapsed label; (ii) a rethinking of the SAE minimal risk condition as a decision to be interpreted, verified, and adjusted against on-scene human authority, rather than treated as a synonym for safety; and (iii) an in-progress survey of research on how autonomous driving systems must be expanded in order to be accessible. Read together, the threads share common themes of encoding disagreement rather than collapsing it, and a need to respond to humans in and around these systems in whichever modality of communication is available. We present these as open problems and invite workshop participants to discuss and develop these ideas with us, grounded in community research and lived experiences.