[AI Competition] Build the best turn-based AI for Athena Crisis

[cpojer]

I want Athena Crisis to have a much stronger AI, and I want the best AI of the genre to be built in the open to inspire every future turn-based strategy game. This is a competition to build a better AI for Athena Crisis that beats the current AI.

All submitted AIs will be ranked based on a set of benchmarks. The prizes are as follows:

• 1st: $2500
• 2nd: $1600
• 3rd: $900

We expect the benchmark to evolve while the competition is ongoing and people are welcome to make contributions to the set of benchmarks we are considering to make sure the AIs can deal with as many situations as possible. The competition is scheduled to run until December 14th, but we may extend the timeline depending on progress and feedback.

Since this task requires extensive knowledge of Athena Crisis, a limited number of Steam Keys for the full game is available for participants. Please reach out to @cpojer (twitter.com/cpojer or christoph@nakazawa.dev) for more information.

Check out the AI Playground.

AI Info

The current AI is called DionysusAlpha and its code can be found there. There are also many AI specific helpers in dionysus/lib.

Rules

• Completeness: Any custom AI must be able to pass all objectives and have as many capabilities as the original AI.
• Performance (Speed): The AI must not be slower at performing actions on a set of benchmarks compared to the current DionysusAlpha AI as of July 29th 2024. Any performance improvements made to existing AI code or libraries that also speed up DionysusAlpha during the competition will only be attributed to your code, but not to DionysusAlpha. This means that speeding up library code is beneficial to you and the codebase, and will not count against you.
  • Note: Any improvements made as part of [Performance] Improve the performance of Radius.tsx by 2x. #9 are considered an exception, and they will speed up DionysusAlpha as well as your AI. On the other hand, any improvements made as part of [AI] Improve AI performance by 5-10x #15 will only count towards your AI, even though it makes the existing AI faster. We care about the baseline of AI performance as of now.
• Stability: The AI must not crash or produce illegal actions.
• Evaluation: In case the AI performs at the same speed with the same outcomes as another AI, it will score based on the fewest turns it takes to win, or otherwise the fewest amount of actions it takes.
• No Hardcoded Map-Specific Logic: AIs should be able to deal with any map setup and must not have map-specific logic hardcoded.
• Choice of Tooling: The AI has to be written in JavaScript and be maintainable. This is not a code-golf competition or challenge about porting code to another programming language.
• Fog: In fog, AIs are receiving more information than regular players. This allows them to make better decisions for gameplay reasons, but could also be considered cheating. DionysusAlpha handles this carefully by hiding information from various algorithms via this.applyVision(map). We suggest using your best judgment about the degree of cheating that your AI is allowed to do. Benchmarks will run fog games with full information and with only visible information so they can be compared.
• Clarifications: We may clarify rules as necessary.

Funding

• We're using Polar.sh to distribute funds.
• You receive the reward once the issue is completed & confirmed by Nakazawa Tech.