Hello,

This has been discussed for previous versions of gazebo, but I could not really find a solution, so I am posting:

Lets say I am making an experiment in gazebo, with a control algorithm. Due to non-deterministic nature of the simulation, I get varying results each time I run the simulation. At first, I thought the variation between results had to be my computer slowing down, especially disk i/o, so I shutdown all the logging, cameras, [I have a top down looking camera, that saves to the disk] - So I shut all the extras off, and rerun my experiment, and I found out that the results still do vary.

The experiment in question uses a small robot, with differential drive plugin, so I plotted the odometry data, to find out, YES, the control algorithm, does indeed produce different outputs, that leads to different simulations.

I have tried stepping ~world_control externally, also to test if any bottleneck in the system was causing variations, but again, even if you wait a long time between steps, it still produces different results, under same circumstances. (a little of course)

Is there any way, that I can get repeatability in my experiments? Are there any parameters for the gazebo sim, that can make the simulation less realistic, but more repeatable.

There has been a discussion about the RNG salt during init, but that was gazebo 1.9, and I could not access the page of discussion, since it was deleted, and it was not also in archive.org.

Any ideas/recomendations/help on the problem much appreciated.

Best Regards, ~C.