Angry AI Birds
It turns out that there are AI competitions for the Angry Birds game over at aibirds.org. The long term goal of the competition is to build an intelligent Angry Birds playing agent that can play new levels better than the best human players, but there are sub competitions to help towards this goal.
It turns out, it's harder then you might think.
Machine learning approaches have been largely unsuccessful. Despite all the successes of deep learning in the past few years, no deep learning based agent has yet entered the semi-final round of our competition. While the best agents play reasonably well, none of them comes close to good human players. The strategy most agents use is to simply find a good shot for the current bird. To encourage agents to plan ahead, we introduced what we call deceptive levels, where the typical greedy strategies that agents use do not lead to good outcomes.
This is why there are competitions for generating Angry Birds levels. A detailed paper on this competition can be found here. The main goal is to generate levels that will break a greedy solution and the paper mentions a few interesting techniques to generate them. Here's a figure from the paper.
The winner from 2021 describes their approach in a paper that can be found here. These are the levels that it generated:
The paper lists the heuristics used in more detail. But I thought the idea of making deceptive levels was an interesting one. Partially because it's a tricky problem, but also because it's a logical building block towards getting a proper bot to play the game. And the "placing a deceptive TNT block to fool bots" made me giggle out loud.
The AI competition also mentions a seperate track that introduce novel challenges that contain previously unseen objects and interactions. You can also participate by writing a bot for their environment, which is explained in more detail here.