COLLISION-FREE VIRTUAL NAVIGATION USING VIEWPOINT SPECIFIC DEPTH INFORMATION(Session2 Virtual Reality)(IWAIT2002)

抄録

Computer animation and virtual reality applications continue to gain popularity and be the focus of increasingly extensive research. However the shortcomings of a virtual environment are numerous and in order to create realistic scenes one must address a lot of practical issues. This paper presents a realistic navigation algorithm for collisionfree agent movement. In the scope of this paper, an agent is considered to be an autonomous, computer assisted or computer controlled entity, which unlike general collision detection and navigation algorithms relies on purely local geometric information of its surroundings to find a path towards a goal position. The approach taken is using depth information directly from the agent's view rendering pipeline to create a repelling force field around obstacles. More specifically the z-buffer of the agent's view is used to estimate a vector field around the obstacles that lie in the agent's path. By making use of the z-buffer the navigation strategy implemented is based solely on information the agent can acquire from its sensors at his present position. This makes the algorithm suitable for autonomous agents in either static or dynamic, unstructured environments, where no a priori information on the topology of the virtual world is known. Furthermore, the algorithm is simple and fast and could be easily extended to real world robots equipped with depth lenses.