Abstract

Abstract. Interacting with humans remains a challenge for autonomous
vehicles (AVs). When a pedestrian wishes to cross the road in front of the
vehicle at an unmarked crossing, the pedestrian and AV must compete
for the space, which may be considered as a game-theoretic interaction in
which one agent must yield to the other. To inform development of new
real-time AV controllers in this setting, this study collects and analy-
ses detailed, manually-annotated, temporal data from real-world human
road crossings as they interact with manual drive vehicles. It studies the
temporal orderings (filtrations) in which features are revealed to the ve-
hicle and their informativeness over time. It presents a new framework
suggesting how optimal stopping controllers may then use such data to
enable an AV to decide when to act (by speeding up, slowing down, or
otherwise signalling intent to the pedestrian) or alternatively, to continue
at its current speed in order to gather additional information from new
features, including signals from that pedestrian, before acting itself.