Efficient architecture for collision detection between heterogeneous data structures application for vision-guided robots

Many collision detection methods exist, each specialized for certain data types under certain constraints. In order to enable rapid development of efficient collision detection procedures, we propose an extensible software architecture that allows for cross-queries between data types, while permitting the time and memory optimizations needed for high-performance. By decomposing collision detection into well-defined algorithmic and data components, we can use the same tree-descent algorithm to execute proximity queries, regardless the data type. We validate our implementation on a path planning problem in which a vision guided humanoid represented by an OBB tree explores a dynamic environment composed of voxel maps.