Section: New Results

Semantically consistent herarchical decomposition of virtual urban environments

Participants : Carl-Johan Jorgensen, Fabrice Lamarche [contact] .

When planning a path in their environment, pedestrians do not consider every detail at once. Instead, people first plan a coarse path, choosing streets to travel to reach their goal. Local decisions such as where to cross a street or on which side to pass by a pole are taken during navigation. In computer science, hierarchical representations of an environment are often used to reduce the computation cost of the planning algorithm. Such a representation also enables smarter navigation behaviours. Indeed, it others the opportunity to delay the local planning until relevant information is available. It also enables a quick recovery from unexpected events, as the high-level path might stay valid even if unexpected events alter the lower-level path.

We proposed a method that automatically generates a three-level hierarchical representation of an informed urban environment. In this hierarchy, each level is a semantically coherent partition of the navigation areas and can be used to plan paths at different levels of abstraction. This representation is used in a path planning process that delays some decisions until relevant information is perceived. This algorithm uses path options to smartly adapt the path when unexpected events occur.