Jørgensen, Jimmy Alison4; Fugl, Andreas Rune4; Petersen, Henrik Gordon4
1 The Maersk Mc-Kinney Moller Institute, Faculty of Engineering, SDU2 The Mads Clausen Institute, Faculty of Engineering, SDU3 SDU Robotics, The Maersk Mc-Kinney Moller Institute, Faculty of Engineering, SDU4 The Maersk Mc-Kinney Moller Institute, Faculty of Engineering, SDU
In this article we present a GPU accelerated, hybrid, narrow phase collision detection algorithm for simulation purposes. The algorithm is based on hierarchical bounding volume tree structures of oriented bounding boxes (OBB) that in the past has shown to be efficient for collision detection. The hierarchical nature of the bounding volume structure complicates an efficient implementation on massively parallel architectures such as modern graphics cards and we therefore propose a hybrid method where only box and triangle overlap tests and transformations are offloaded to the graphics card. When exploiting coarse-grained parallelism in grasping and stacking simulations, requiring all-contacts resolu- tion, a performance gain of up to 7x compared to the collision detection package PQP is obtained.