Generative Pre-trained Traffic Control: using LLMs for air traffic control
This code is written as a plugin for the BlueSky ATC simulator.
Running install.sh will place the necessary files in your BlueSky plugin directory if bluesky-simulator is installed as a package.
For other installation methods, you will have to manually place the files in the correct directory.
Edit the BlueSky configuration file to enable the plugin.
You will need a .env file specifying your LLM API access key and related information.
The plugin currently expects the use of the University of Michigan ChatGPT API.
In the simulator, use gptc on and gptc off to toggle the activity of the plugin.
The default state is off.
To collect statistics, the start_gptc_scenario <filename> command allows for the running of the same scenario file as many times as needed. Just add this command as the last line in your scenario file, e.g.
00:07:57.00>start_gptc_scenario 740_heading_repeat.scn
and the plugin will write out separation violation information to scenario_results.txt.
