AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |
Back to Blog
Stratego game data ai4/7/2023 DeepNash beats existing state-of-the-art AI methods in Stratego and achieved a yearly (2022) and all-time top-3 rank on the Gravon games platform, competing with human expert players. The Regularised Nash Dynamics (R-NaD) algorithm, a key component of DeepNash, converges to an approximate Nash equilibrium, instead of 'cycling' around it, by directly modifying the underlying multi-agent learning dynamics. DeepNash uses a game-theoretic, model-free deep reinforcement learning method, without search, that learns to master Stratego via self-play. For these reasons, Stratego has been a grand challenge for the field of AI for decades, and existing AI methods barely reach an amateur level of play. Episodes are long, with often hundreds of moves before a player wins, and situations in Stratego can not easily be broken down into manageably-sized sub-problems as in poker. Decisions in Stratego are made over a large number of discrete actions with no obvious link between action and outcome. An AI called DeepNash, made by London-based company DeepMind, has matched expert humans at Stratego, a board game that requires long-term strategic thinking in the face of imperfect information. This popular game has an enormous game tree on the order of $10^$ nodes). From a report: Another game long considered extremely difficult for artificial intelligence (AI) to master has fallen to machines. Stratego is one of the few iconic board games that Artificial Intelligence (AI) has not yet mastered. In the future I will allow the player to place his pieces as he wants, though.We introduce DeepNash, an autonomous agent capable of learning to play the imperfect information game Stratego from scratch, up to a human expert level. (Note: in the current version, pieces are placed randomly on the board for both players. The question is: are there any patterns that can be used as the basis of an algorithm to figure out where the enemy's flag might be? When human players play against each other, what are their tactics to find out the flag's location? It will jump onto opportunities to capture pieces, but it never "plans" to move anywhere, and it never takes the Flag, since it thinks that the Flag might as well be a bomb. My problem is that the AI is currently extremely defensive. If an unknown piece moves up to the General, the latter will flee, as the unknown piece might be a Marshal. If the spy is no longer alive, the Marshal will only be afraid of non-moving pieces, since those could be bombs. For instance, a Marshal will not capture a piece if there's an undiscovered piece next to it, since this latter piece could be a spy. Pieces that have not yet been revealed are also taken into consideration. The AI remembers ranks of pieces that have already been revealed, and will capture or flee depending on this. The thing is, right now my AI works pretty well already. (My current work assignments involve no programming.) This info is not entirely relevant though. I am currently programming a Stratego game in JavaScript, simply because I wanted to learn to use JavaScript while spending some time to keep my programming skills up to par. 2 How to Play When we build AI’s, we are trying to solve a specific problem, but the algorithms and strategies used can be generalized.
0 Comments
Read More
Leave a Reply. |