Interactive navigation of humans from a game theoretic perspective

Humans are more successful in planning collision free, continuous trajectories through populated environments than any motion planning algorithm so far. This is due to the fact that they consider the conditionally cooperative, interactive behavior of the surrounding persons, for example the possibility of mutual avoidance maneuvers. In this paper, interaction during navigation is regarded from a game theoretic perspective and the concept of Nash equilibria is applied to analyze human motion. In contrast to other methods, the game theoretic approach does not necessarily rely on learning the interaction itself and is extendable. Our approach is based on human motion data that is captured during experiments. Two hypotheses are verified: for one thing, interaction exists during human navigation, for another thing, the mutual avoidance behavior of humans can be modeled with the theory of Nash equilibria in non-cooperative games. This knowledge can be used to enhance existing motion planning algorithms for autonomous robots.