ClashRoyaleAi: an open-source, deterministic Clash Royale simulator for RL, with recurrent PPO, lookahead search and expert iteration [P]

Excerpt

Clash Royale is a real-time, two-player, partially observed card game with no API. We rebuilt its battle engine as a headless C++17 simulator (pybind11) so it can serve as an RL environment. **Environment** - Every card in the game as of July 2026 (132 cards + 41 Evolutions), deterministic and seedable - ~10 ms per full match on one laptop core; `snapshot()` forks any state in 7–20 µs (lookahead, paired A/B tests) - Gymnasium API. The 13,977-float observation includes the opponent's played-card