This repo contains a proof-of-concept implementation of the transformer protocol described in the (under submission) paper "Automatic Distributed Multi-Agent Coordination of Single-Agent Robot Controllers", as well as a simulator to run said implementation and a visualization generator for the results of the simulator.
You'll need the following dependencies (all Python modules; install as you prefer): coloredlogs,
icecream, verboselogs, numpy, fire, matplotlib. This is Python 3 code.
You can run python simulate.py --help to see documentation on usage of the simulator; a good
example invocation is python simulate.py --num-agents=5 --verbose to generate a random scenario
with 5 agents and use verbose logging.
There's also python visualize.py --help; a corresponding example here might be python visualize.py scenario_foo.json output_bar.json --output-file=foobar.mp4 to load some scenario
scenario_foo.json, some simulation trace output_bar.json, and save the resulting visualization
to foobar.mp4.
This is research-grade code. As such, it comes with no support or guarantees, and is likely buggy. Fixes for bugs will be added as they are discovered and as time permits.