(Slight) Randomness in results

[ChristianWesterdahl]

Using the algorithm on a custom stereo camera and IMU setup, we have noticed that each run on the algorithm on the same data will produce a (slightly) different path. Even though it is the same data and parameters we use when playing the rosbag and the algorithm. We make sure to run the sim_core node and use the --clock argument for the rosbag.

Any ideas as to where this inherent randomness originates? My first guess was the feature extractor, but after looking into the documentation, it should be completely deterministic. I cannot find anything in the paper, and as far as I can see, all the “random” modules in the code are used to generate tests.

Is it simply a consequence of running the algorithm in ROS?