Methodology
We designed eight benchmark tests to evaluate the accuracy and the speed of contact simulation from the physics engines. The implementation of the simulation instances was developed based on the software framework that we created using user-level API of the physics engine to enable to make uniform test environments.
Our benchmark software framework features are as follows:
- simulation related functions
- graphical user interface
- CPU time checking, data collection, and visualization
We characterized the speed of the simulation by a real-time factor(RF) that is simulation time divided by raw CPU time. The accuracy metric for each test is described on the test pages.
For representation of body collision shapes, we only used primitive collision shapes(sphere, cylinder, box, capsule) since collision detection of mesh object is considerably expensive. Note that we set all parameter that may cause unrealistic and unpredictable physical interaction into zero; they are ERP(Error Reduction Parameter), CFM(Constraint Force Mixing) and damping parameters.
The details of implementation for each physics engine is stated here.
Test Machine
- Intel Core i7-7700HQ CPU @ 2.80 GHz processor
- 15.4 GiB memory
- Ubuntu 16.04 LTS 64-bit