Sleep in het vak om omhoog en omlaag te scrollen. De tekst loopt mee met de paginabreedte.
| Link | Abstract |
|---|---|
| https://github.com/bcollazo/catanatron | Popular, fast, and feature-rich Python Settlers of Catan simulator with strong built-in AI players (including MCTS/heuristic bots). Excellent foundation for experimenting with new agents, data analysis, or reinforcement learning—widely used as a benchmark in the Catan AI community. |
| https://settlers-rl.github.io/ https://github.com/henrycharlesworth/settlers_of_catan_RL |
In-depth project using PPO (Proximal Policy Optimization) for deep RL in Catan, with a custom simulator, action/observation spaces, and forward-search improvements. Great for studying advanced RL techniques applied to imperfect-information games. |
| https://github.com/kvombatkere/Catan-AI | Full multiplayer Catan framework in Python supporting RL + MCTS agents. Solid for building, testing, and comparing reinforcement learning or search-based bots in a complete game environment. |
| https://github.com/akrishna77/CS7641-QSettlers | Deep Q-Learning (DQN) approach to Catan in a Java-based framework with rule-based baseline agents. Useful academic-style project for understanding value-based RL in multi-player, stochastic games like Catan. |
| https://justinasher.me/catan_ai | Neural network trained via supervised self-play to reach intermediate-level Catan play. Clean implementation with custom game logic—ideal for exploring self-play pipelines without heavy RL overhead. |
| https://github.com/HenrikFredriksen/sub-catan | Simplified multi-agent RL environment (AECEnv/TMARL compatible) based on a stripped-down version of Catan. Perfect for quick prototyping of cooperative/competitive multi-agent algorithms or benchmarking in a lighter setting. |
| https://github.com/josefwaller/PyCatan2 | Solid, lightweight Catan simulator—great for basic game logic, bots, or extensions without heavy AI focus. Successor to the original PyCatan; provides core rules and flexibility for custom variants or simple agents. |
| https://github.com/striest/catan_bot | Python bot specializing in optimal initial settlement placements using AI techniques. Great for strategy analysis or as a starting point for placement heuristics in larger simulators. |
| https://github.com/williamaredal/Catan-Simulator | Straightforward tool for batch simulations of N games and analyzing stats/strategies. Useful for data-driven insights into win rates, resource probabilities, or testing rule tweaks. Full-stack: React frontend for visualization + Python backend for simulations. |
| https://github.com/adammann52/catan | Clean Python implementation of Catan with a GUI (using Pygame or similar) for human play. Supports 3-4 players; good foundation if someone wants to add AI later or visualize games. |
| https://github.com/jdmonin/JSettlers2 | The classic, full-featured open-source Settlers of Catan implementation in Java — complete with multiplayer networking, built-in bots, and expansion support. Widely referenced as a benchmark in Catan AI research; great for studying robust game engines, even if you're working in Python. |
| https://github.com/avtnl/Catan_Gen3_v037 | Anthony's public Catan engine work in Python: a Gen2 line (Catan_Gen2_v045) and a Gen3 line (v010b, v016_experiment, v032-v035). Later Gen3 versions add player trading (TwP) and development cards. Companion site: https://www.catan-and-ai.org Gen3 v037: pygame GUI for mixed human/AI (default human vs 3 AIs); run_headless.py batch sims; replay_catan_game.py steps a chosen game from batch_runs/gNNN. |
| https://github.com/cullback/canopy | Rust PUCT/Gumbel-AlphaZero framework (chance nodes, SO-ISMCTS, batched GPU inference). Flagship example is 1v1 Catan; README claims nexus-v3 is the strongest public 1v1 Catan agent, with a web analysis board and colonist.io CDP play. Stars: 5. Last push: 2026-04-16. |
| https://github.com/nickita-khylkouski/catan-zero-public | Expert-iteration / Gumbel-AlphaZero stack for 2-player no-trade Catan. Coherent public-belief MCTS, information-set invariance, fleet training (H100/B200), promotion vs champion B12. Python plus native Rust search. Stars: 1. Last push: 2026-07-21. |
| https://github.com/Eli6th/catan-rl | Zero-alloc Rust engine (~88 ns/step), PyO3 batched env, masked PPO self-play, AlphaZero-lite search. Reports 82% vs heuristic opponents in 4-player first-to-7. Stars: 1. Last push: 2026-06-29. |
| https://github.com/rli8145/alphahex | Standalone 1v1 Catan vs AlphaZero-style MCTS (heuristic plus neural value/policy, PUCT). Playable at https://alphahex.vercel.app/. Evolutionary weight training and NN self-play. Stars: 1. Last push: 2026-08-24. |
| https://github.com/Samffprice/bettafish | AlphaBeta, MCTS, and AlphaZero on a fast bitboard Catan engine. Stars: 0. Last push: 2026-02-24. |
| https://github.com/Wind-Eagle/CataNN | MCTS agent with neural network value estimation, written in Rust. Stars: 0. Last push: 2026-07-03. |
| https://github.com/JdHeer/AlphaZero-Catan | AlphaZero on Catanatron: policy/value net, MCTS, self-play, iterative training. Stars: 0. Last push: 2026-01-31. |
| https://github.com/AlgoCatan/RLCatan | Catanatron fork: PPO + self-play + curriculum + Elo league. Live arena at https://rlcatan.vercel.app/. LLM-generated move explanations. Stars: 3. Last push: 2026-07-09. |
| https://github.com/nogulong/rust-catan-rl | Fork of Swynfel/rust-catan with end-to-end RL agents; eval vs JSettlers (paper in prep). Stars: 3. Last push: ~2026-04-30. |
| https://github.com/mcrco/catanrl | Broad Catan RL suite on vendored Catanatron: DAgger, single-agent PPO, MARL PPO with centralized critic, AlphaZero, PufferLib wrappers, hidden-info beliefs / IS-MCTS. Stars: 0. Active. |
| https://github.com/edtireli/catan-sim | Full-rules simulator plus 1.5M-param PPO self-play, action masking, React UI, spectator/replays, TensorBoard. Stars: 0. Last push: 2026-04-20. |
| https://github.com/PeterLP123/catanatron-1v1 | GPL fork of Catanatron for Colonist-style 1v1 (15 VP). Gymnasium + masks, BC, DAgger, MaskablePPO, league eval. Stars: 1. Last push: 2026-07-21. |
| https://github.com/Gallind/Genetic-NN-Catan | Genetic algorithm evolves NN weights as a Catanatron position evaluator; NumPy inference, bootstrap from MC/AB labels, curriculum + hall-of-fame. Stars: 1. Last push: 2026-05-24. |
| https://github.com/BKaperick/CatanLearning.jl | RL players on Catan.jl: TD learning, random-forest value from ~500k games, linear/search hybrids (leaderboard in README). Stars: 2. Last push: 2025-10-15. |
| https://github.com/BKaperick/Catan.jl | Julia engine for humans plus scripted AI; points to CatanLearning.jl for ML. Stars: 6. Last push: 2025-10-23. |
| https://github.com/OfficialLahusa/CatanSim | Simulator plus ML and handcrafted agents, parallel dataset generator, feature-vector train/val/test split. Stars: 2. Last push: 2026-08-21. |
| https://github.com/Atharva-Thakur/PPO-based-RL-agent-for-catan | Masked PPO (PyTorch) on catanatron_gym (obs 614, actions 290). Stars: 0. Last push: 2026-02-21. |
| https://github.com/GSJProgo/RL-Catan | Custom 2-player env (no P2P trades); A3C (~87% vs random) and DQN in PyTorch. Stars: 1. Last push: 2024-01-07. |
| https://github.com/Gregory-Eales/gym-catan | OpenAI Gym catan-v0 with feature-plane/vector observations, tunable rewards, GUI. Older but a dedicated Gym Catan env. Stars: 6. Last push: 2020-09-30. |
| https://github.com/nicorising/catan-ai | Catanatron agents compared: MCTS, TD, genetic algorithm, DQN (construction-focused, no trades). Stars: 1. Last push: 2025-08-20. |
| https://github.com/piktiv/settlers-of-catan-thesis | Thesis code: Dueling Double DQN + PER (D3QN+PER) on a custom Catan env. Stars: 1. Last push: 2022-06-20. |
| https://github.com/nbelle1/strategy-game-agents | Official code for Agents of Change / HexMachina (arXiv:2506.04651): LangGraph Analyst/Orchestrator/Coder evolve a compiled Catanatron player (reported 54% vs AlphaBeta). Stars: 17. Last push: 2025-10-23. |
| https://github.com/OguroHayato/hexmachina-repro | Reproduction of Belle et al.: paper artifact measures 45.8% not 54.1%; their best 38.6%. Adds 4-player transfer, opponent curricula, per-turn LLM baseline (0/30). Stars: 0. Last push: 2026-08-07. |
| https://github.com/SoumilRathi/catanbench-v0 | LLM tournament on Catanatron (GPT/Claude/Gemini/OpenRouter), JSON decisions, strategic prompts. Stars: 5. Last push: 2025-08-26. |
| https://github.com/taziksh/catan-llm | Frontier LLM bench on Catanatron, then SFT to DAgger to DPO to GRPO on Qwen3.5-9B via Prime Intellect verifiers. Stars: 2. Last push: 2026-08-21. |
| https://github.com/TFC1224/catan-rl-llm | Qwen3-8B plus Catanatron: SFT, GRPO, VF-guardrails, tools. VF-SFT 76% vs WeightedRandom (50 games); hybrid tools+VF 100% (small n). Stars: 2. Last push: 2026-08-10. |
| https://github.com/vmmadathil/catan-bench | Prompt-only 4-LLM table on Catanatron (scratchpad, 1-for-1 trades). Companion to the Feb 2026 Can LLMs Play Catan? write-up. Stars: 0. Last push: 2026-02-10. |
| https://github.com/alexmaraval/catan-bench | OpenAI-compatible LLM players on Catanatron; Streamlit dashboard, Elo/Bradley-Terry snapshot (Apr 2026). Stars: 0. Last push: 2026-04-14. |
| https://github.com/infoFiets/llm-catan-arena | Catanatron plus llm-game-utils: text and MCP tool-calling LLM players, Elo, cost tracking. Stars: 1. Last push: 2026-02-18. |
| https://github.com/ACosmicCake/Catan-LLM-AI-Arena | Pygame arena: ChatGPT/Gemini/Claude/Deepseek vs heuristic; shows LLM thoughts. Stars: 1. Last push: 2025-06-25. |
| https://github.com/JordiCan/catan_ai_agents | PyCatan-based: heuristic agent plus hybrid LLM (Poligpt/Ollama/Bedrock) and benchmarks. Stars: 1. Last push: 2026-04-27. |
| https://github.com/vasco-s-pereira/CatanAgent | AASMA 2025/26: Nash-guided LangChain workflow vs vanilla LLM vs heuristic for Catan trades on Catanatron. Stars: 0. Last push: 2026-06-02. |
| https://github.com/Algorhythm-sxv/monte-catano | UCI-like CLI MCTS engine in Rust, self-play demo, SPRT match runner. No GUI. Stars: 13. Last push: 2026-06-12. |
| https://github.com/hubertbalos/socAI | Parallel MCTS with move pruning and virtual-win heuristics; Pygame visualization. Stars: 1. Last push: 2024-09-24. |
| https://github.com/kindalime/catan-agent | CS degree project: information-set MCTS (IS-MCTS) Catan agents. Stars: 0. Last push: 2023-05-03. |
| https://github.com/Sharperrr/Catan_MCTS | MCTS (plus random/rules agents) on a modified NatakAPI. Stars: 0. Last push: 2025-05-22. |
| https://github.com/maxyurk/settlers_of_catan | Expectimax, Monte Carlo, and other AI/ML players. Inactive since 2019 but still a high-star leftover. Stars: 18. Last push: 2019-11-30. |
| https://github.com/0xBrsm/HexN | PPO self-play, expert iteration and a resumable training loop for a graph-native network that reads the hex/vertex/edge board graph directly and emits its policy per node (legality as a node property, not a flat mask). Sibling of HexSet. Stars: 0. Last push: 2026-09-23. |
| https://github.com/0xBrsm/HexSet | Catan-rules research platform: NumPy engine with player trading, ONNX inference, browser UI, HTTP and MCP interfaces, Gymnasium and PettingZoo adapters, batched training envs; Catanatron AlphaBeta seatable as reference. Training lives in HexN. Stars: 0. Created: 2026-10-02. |
| https://github.com/Jaime888888/jaime-catan | Rust AlphaZero-style self-play + MCTS for four-player Catan, parallel CPU game workers feeding batched GPU inference; arena evaluation. USC EE 451 (spring 2026) project with report. Stars: 0. Last push: 2026-09-17. |
| https://github.com/osolmaz/catanarchy | Deterministic Catan simulator for agent research: bounded negotiation rounds (offers, counteroffers, messages, promises, atomic trade settlement), web app with replay, agent harness for scripted or model-driven games. Stars: 2. Last push: 2026-09-14. |
| https://github.com/rhayenga/ML-Game-Theory-Strategy-Game-Engine | Local coaching engine: C++20 rules + strategy engine, MCTS top-3 move advice, PyTorch value net trained from C++ self-play features, browser UI and end-of-game recap. Stars: 1. Last push: 2026-09-11. |
| https://github.com/restachoc/catan | Fast Rust bitboard engine (flat 253-action space) with PyO3 parallel VecEnv; PPO and AlphaZero trainers using a board-structured GNN that generalises across layouts; FastAPI web UI to play bots and watch replays. Stars: 0. Last push: 2026-10-04. |
| https://github.com/SpideyZac/catan-ai | Bot that actually trades: Rust engine with full player-trading protocol (~2-3M actions/s/core), entity-transformer policy with topology-aware attention, PPO self-play vs a league of past versions; browser game incl. watching four AIs negotiate. Stars: 0. Last push: 2026-10-03. |
| https://github.com/TSVRN9/settlers_of_catan_rl | Depth-2 expectimax over a learned win-probability net (expert iteration on rollout-labelled states) with trade-offer search. README claims 85.0% of 2,000 games vs three JSettlers 2.6.10 robots and 93.9% of 1,000 vs three Catanatron AlphaBeta. Static site to play it. Stars: 0. Last push: 2026-10-01. |
| https://github.com/Jonathanhsiao123/Catan-RL-Agent | PPO actor-critic self-play on Catanatron (~290 masked actions), pool of frozen past checkpoints, potential-based reward shaping; trades priced by the calibrated value function; browser UI vs humans. Stars: 0. Last push: 2026-09-06. |
| https://github.com/Al2800/catan-llm | Train open-weights Qwen3.5-9B to play Catan in Catanatron: QLoRA SFT on millions of bot decisions, then GRPO with engine-verifiable rewards; gated bot-ladder evaluation. In Phase 2 (SFT). Stars: 0. Last push: 2026-08-31. |
| https://github.com/mustafaTokmak/catan-llm-arena | Any OpenRouter model seated on Catanatron with per-seat cost accounting; first 50-game run (Aug 2026), deepseek-v4-flash won 19, but the author notes models could not see the board so it is not yet a skill claim. Stars: 0. Last push: 2026-08-23. |
| https://github.com/srcolinas/llms-catan | LLM agents of increasing capability playing a Catan reskin (Teyuna – The Lost City) through the author's own game server, traced with Langfuse. Stars: 0. Last push: 2026-09-16. |
| https://github.com/justinmay/CatanBench | JSON HTTP API for autonomous agents to play Catan: register, read private state, choose from legalActions, submit versioned idempotent actions. Stars: 0. Last push: 2026-08-05. |
| https://github.com/mcrco/cppanatron | C++20 rewrite of the Catanatron engine (Codex-written) with a C ABI for Python/PufferLib RL, differential parity tests vs pinned Catanatron. Companion to mcrco/catanrl. Stars: 0. Last push: 2026-08-05. |
| https://github.com/Cheekywnl/catan-ai-lab | Four-player engine with hidden-card beliefs, opening economics, continuation search and win forecasts (runs in browser). Tactical v4 won 215/800 (26.9%) vs three v3 bots. Paused 12 Sep 2026. Stars: 0. Last push: 2026-09-12. |
| https://github.com/mfradley/catan-ai | Pure-Python full-rules engine incl. multi-round player negotiation, pygame GUI, and a 2-player self-play masked-PPO bot (sb3-contrib) with hidden-information-preserving observations. Stars: 0. Last push: 2026-08-24. |
| https://github.com/guidodinello/catan | Base-game rules engine as a Monte Carlo / ML research platform: opening placement, position win probability, build orders, RL vs heuristic agents. Stars: 0. Created: 2026-09-06. |
| https://github.com/TamSeid/catan-ai | Stdlib-only full-rules engine with weighted heuristic play-style presets and MCTS rollouts over opening settlements; batch self-play harness for strategy comparisons. Stars: 0. Last push: 2026-08-26. |
| https://github.com/seanpatrickmay/AlphaSettler | Very new (no README yet). Description: Superhuman Settlers of Catan AI: fast Rust engine, self-play search. Worth rechecking next month. Stars: 0. Last push: 2026-10-02. |