til-25-finals

Congratulations on making it through to the BrainHack TIL-AI 2025 Qualifiers!

Your autonomous reconnaissance bots are now operationally ready. As they navigate through the RL environment, when they encounter special missions, they will initiate your other mission tasks (e.g. ASR, CV, OCR) to be processed in the background and returned to HQ when a result is found. This is facilitated by your model orchestration server in finals, which should call your other model containers appropriately then pass the result back to our competition server. To let you test whether your solution works end-to-end, we have provided you a version of our competition server (HQ) for you to use when testing everything on your GCP instance.

There are also some slight changes to the til-25-environment, and it is recommended you ensure that your existing RL agents work (and possibly are re-trained) with the changes to the environment. Most notably:

• A bug has been fixed in the spawning locations of the agents. They now correctly alternate spawning across each of the corners of the map, rather than just the top left as a Scout and a subset of the rest of the locations.
• An infinite loop in arena generation has been patched.
• A flag has been exposed in the info dictionary returned to indicate whether a special mission has been initiated.
• The step observation component has been modified to have a maximum value of NUM_ITERS+1 instead of NUM_ITERS to reduce the likelihood of off-by-one errors, and is now also manually cast to a np.uint8.

Setup

Init and update all submodules (til-25-environment within the test_competition_server directory).

git submodule update --init

Create a new .env file copied from the .env.example file. Also create a directory called artifacts to store testing artifacts from Docker so you can review them later:

cp .env.example .env
mkdir -p artifacts

Also make sure that your data directory (either novice or advanced) is mounted in your home directory. If it's not, you should be able to mount it with the following:

mkdir -p $HOME/$TRACK && sudo mount $HOME/$TRACK

Submitting for finals

Submit the final versions of all your models using the til submit command, using the naming convention {TEAM_NAME}-{TASK}:finals. For example, if your team name is team-ryan and you're submitting your asr model for finals, the name would be team-ryan-asr:finals.

til submit team-ryan-asr:finals
til submit team-ryan-cv:finals
til submit team-ryan-ocr:finals
til submit team-ryan-rl:finals

Run

To test everything working together, run the test.sh script, which will:

1. Build your model orchestration server from the finals/ directory, and push it to Artifact Registry.
2. Run docker compose up, testing the full setup end-to-end with a mock competition server.

bash test.sh

It should take a couple of minutes to run because it waits for 2 seconds for all the RL results to return. If you know your RL agent runs substantially faster than that (as the overwhelming majority of the submitted RL agents do), then feel free to modify the RL_TIME_CUTOFF value in src/constants.py so your test runs faster.

If everything works without errors, hooray! We'll see you at the IRL finals at MBS on June 11th and 12th <3