Real-time house-collapsing simulations based on performance evaluation of physics engines

Abstract

House-collapsing simulation using a physics engine is an effective method for acquiring structural data regarding collapsed houses with the aim of understanding the properties of destroyed or disordered structures for designing and operating rescue-robots. However, the simulation needs a lot of time since a house model consists of a large number of rigid bodies and joints. In order to find an appropriate configuration of computer hardware and software for accelerating house-collapsing simulations, this study evaluates the performances of four major physics engines, namely Open Dynamics Engine, Bullet Physics Library, PhysX 2.8.1 and PhysX 3.4, by comparing the processing time about two sample structures including only rigid bodies or rigid bodies constrained by joints. Results of the experiments show that the use of multi-core CPU and GPU, especially high-speed GPU, on PhysX 3.4 has the best performance since it can process a large number of rigid bodies and joints in parallel. Based on the results, an existing simulation system has been improved and the collapsing process of one-house, which consists of about 7,500 rigid bodies and 15,000 joints, can be simulated in real-time using multi-core CPU and GPU. Moreover, this study estimates the size of GPU memory which is required for simulating the large-scale field on PhysX 3.4 to enlarge the scale of simulation. Consequently, the collapsing process of thirty-houses which includes interactions between collapsed houses can be simulated.