AI Agents for the Dhumbal Card Game: A Comparative Study

TL;DR

The paper tackles AI design for Dhumbal, a multiplayer imperfect-information card game, by formalizing its rules and evaluating rule-based, search-based, and learning-based agents. Through 1024-round within-category tournaments and a cross-category championship, it shows a simple Aggressive heuristic achieving the highest win rate and positive economics, outperforming more computationally intensive ISMCTS and PPO approaches. These findings highlight the value of domain-specific heuristics in certain imperfect-information settings and provide a reproducible framework and open-source code for future work and cultural preservation. The study thus contributes to AI research in imperfect-information games and supports digital preservation of traditional games by delivering practical benchmarks and methodologies.

Abstract

This study evaluates Artificial Intelligence (AI) agents for Dhumbal, a culturally significant multiplayer card game with imperfect information, through a systematic comparison of rule-based, search-based, and learning-based strategies. We formalize Dhumbal's mechanics and implement diverse agents, including heuristic approaches (Aggressive, Conservative, Balanced, Opportunistic), search-based methods such as Monte Carlo Tree Search (MCTS) and Information Set Monte Carlo Tree Search (ISMCTS), and reinforcement learning approaches including Deep Q-Network (DQN) and Proximal Policy Optimization (PPO), and a random baseline. Evaluation involves within-category tournaments followed by a cross-category championship. Performance is measured via win rate, economic outcome, Jhyap success, cards discarded per round, risk assessment, and decision efficiency. Statistical significance is assessed using Welch's t-test with Bonferroni correction, effect sizes via Cohen's d, and 95% confidence intervals (CI). Across 1024 simulated rounds, the rule-based Aggressive agent achieves the highest win rate (88.3%, 95% CI: [86.3, 90.3]), outperforming ISMCTS (9.0%) and PPO (1.5%) through effective exploitation of Jhyap declarations. The study contributes a reproducible AI framework, insights into heuristic efficacy under partial information, and open-source code, thereby advancing AI research and supporting digital preservation of cultural games.

Figures (7)

• Figure 1: Pairwise player performance comparison using Cohen's d effect sizes. Positive values (red) indicate the first player outperformed the second; negative values (blue) indicate the opposite. Asterisks denote significance levels: * $p < 0.05$, ** $p < 0.01$, *** $p < 0.001$.
• Figure 2: Performance comparison of rule-based agents across three key metrics: (a) win rates with 95% confidence intervals, (b) economic performance, and (c) Jhyap success rates. All metrics are based on 1024 simulation rounds.
• Figure 3: Tournament performance of MCTS and ISMCTS agents showing (a) win rates with 95% confidence intervals from 1024 rounds and (b) cumulative win progression throughout the tournament.
• Figure 4: Training dynamics of PPO and DQN agents across 1024 episodes showing (a) reward progression with 100-episode moving average, (b) cumulative win rate convergence during training, (c) episode length indicating game efficiency, and (d) learning stability measured by 100-episode rolling standard deviation.
• Figure 5: Tournament performance of PPO and DQN agents showing (a) win rates with 95% confidence intervals from 1024 rounds and (b) cumulative win progression throughout the tournament.