Page 1 of 1

Performance vs. pybullet

Posted: Thu Feb 18, 2021 2:39 am
by locked21
Hello. I will be working on a reinforcement learning project for the next days. The way I intend to use bullet is to run a simulation of a small car (like one of those from the examples) for about 5000 simulation steps, then resetting the robots to an initial position and repeating. Every 10~50 steps I'll be reading ray scans results and taking 1~4 images from a mounted camera.

Since this project will require a huge amount of computation I want to know if it's worth changing the code from pybullet to cpp bullet. Do we have any estimates on the gains of using cpp bullet, with enhancements like CPU or GPU (I have a GTX960) parallelization, versus pybullet? I know I'm asking a hard or even subjective question, but any amount of information may help me in this decision. Thanks