Parallel and Hardware Implementations of Collision Detection Algorithms

• Main Author: Fenton-May, Tristam
• Published: University of Oxford 2004
• Subjects: 004
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.491445

In recent years many graphics processes, which have traditionally been implemented as software running on general purpose hardware, have re-emerged as low cost application specific hardware. This hardware is now available on many desktop machines to accelerate the rendering of graphical scenes, and thus allowing high quality real-time visualisation. Another aspect of simulation which is computationally expensive is collision detection. This document looks at existing algorithms for collision detection, and how they might be implemented using special purpose hardware to improve their efficiency. A general framework is described that allows multiple collision processing algorithms to be brought together to build a distributed collision farm. An implementation of this system was built using desktop PCs connected via a LAN. The farm is capable of processing collisions of up to a million polyhedra in real time using 30 (now outdated) machines. To achieve even greater speed a hardware implementation is described. The widely used Sweep & Prune algorithm forms part of many software collision systems, and this hM been implemented in hardware using a new type of hardware sorting array. A physical implementation has been built using VHDL and an Altera programmable logic device. The results show that a system using similar number of gates to a 486 (Le., about 1% of the size of a modern graphics processor) can perform this task faster than a processor farm comprised of 14 Intel Pentium Pro CPUs and a purely software implementation.