Modélisation topologique et sémantique de l’environnement

Abstract

3D environments have appeared at first devoid of mobile entities and purely geometric. Virtual characters have been subsequently added, creating a pseudo-life within these environments. These characters were an animation; they followed a path predetermined by their respective creators. As and when these characters are endowed with certain autonomy, they were now able to analyze briefly the outside world to choose the best strategy. These humanoids are on time today able to move in a complex environment, to save themselves from each other or even to capture some semantic environment. This personal navigation is essential to the realism of simulations. To increase the realism we focused on the problem of modeling the environment in which these characters are independent. To facilitate navigation of the latter, it is important to capture the semantics of this environment. Indeed, the world around us, our behavior is modeled on our site analysis in which we sail. We do not adopt the same attitude when we travel on a sidewalk when we cross a street with heavy traffic. Our speed is continuously adapted to the environment and varies gradually. Thus, it is reduced when approaching an obstacle, a staircase or a turn. We propose a unified model of the environment that can take into account the topology and semantics of the premises. The topological representation of the environment is 2D1/2.It is obtained automatically by using an exact decomposition of navigable surfaces in convex cells, rather than a breakdown in a grid considered too rough. To the topological representation we associate with certain areas of the surface the type of area (street, grass, doors, stairs, etc ...). This information allows the humanoid to know the place where it is located and thus allows it to adapt its behavior.