Abstract

Autonomous vehicles (AV’s) are appearing on roads, based on standard robotic mapping and
navigation algorithms. However their ability to interact with other road-users is much less well understood. If
AVs are programmed to stop every time another road user obstructs them, then other road users simply learn that
they can take priority at every interaction, and the AV will make little or no progress. This issue is especially
important in the case of a pedestrian crossing the road in front of the AV. The present methods paper expands the
sequential chicken model introduced in (Fox et al., 2018), using empirical data to measure behavior of humans in
a controlled plus-maze experiment, and showing how such data can be used to infer parameters of the model via
a Gaussian Process. This providing a more realistic, empirical understanding of the human factors intelligence
required by future autonomous vehicles.