FURI | Spring 2020

Socially Attentive Autonomous Vehicles

The computational cost of determining the action sequence in high traffic areas such as changing lanes on a highway or merging into lanes can be computationally costly. Rather than computing the Nash Equilibrium on all the cars in the environment, suppose the Nash Equilibrium was computed pairwise rather than on the whole environment. This would not only reduce the complexity of the problem, but given the assumption of a perfect pairwise controller, the lightweight attentive model would be the only thing needed to scale different interactions.