juloberno
commented
2 years ago Thanks for pointing this out. I checked it in the example.
The paper uses the ConfigurableScenarioGeneration, which allows setting a parameter for the random seed, which achieves reproducible scenario sets for a given parameter file. The ConfigWithEase scenario generation used in the examples does not include this option. A workaround is to add
numpy.random.seed(some_arbitrary_seed)at the beginning of the example's python file.
To your second question: You find all files generated during the build in the bazel build file folders, e.g. for the intersection example you have ".../bark/bazel-bin/bark/examples/intersection.runfiles/bark_project/bark/core.so".
Repeatability is mentioned In the docement "BARK: Open Behavior Benchmarking in Multi-Agent Environments". “Experiments in BARK shall be fully reproducible, independent of the frequency at which the simulation runs.” However, when I run the example by run "bazel run //bark/examples:intersection", I find that it runs multiple times and the execution results are different. What causes this difference？ Besides, I didn't find the bag bark.core. "from bark.core.models.behavior import *", I wonder how the core bag can be found. I would appreciate it if you could help me answer these two questions！