DVSA-Simulator

DVSA-Simulator

A high-performance simulation framework for turn-based tactical games, designed with AI experimentation in mind.

Deep Strategical Game

Complex skill interactions, unit classes, and terrain effects allow for rich, long-horizon planning—perfect for evaluating decision-making agents.
Unlike deterministic board games like Chess or Shogi, this system includes probabilistic elements (e.g., hit rates, skill triggers, HP dynamics), making planning under uncertainty significantly more challenging.

Reinforcement Learning Friendly API

Designed with Deep Reinforcement Learning in mind, the simulator cleanly separates agent and environment logic. The environment exposes a gym-like API with step/reset interfaces, and supports both discrete and parameterized action spaces. External control is possible via Python FFI, enabling seamless integration with existing RL libraries (e.g. Stable-Baselines3, RLlib, CleanRL). Useful for simulating turn-based decision-making under uncertainty in a structured and reproducible way.

High Memory Efficiency

Carefully designed data layout ensures high cache locality, reducing memory access latency during simulations. The simulator avoids unnecessary allocations and leverages contiguous memory structures for fast iteration and low-overhead rollouts. Ideal for large-scale simulations where throughput is critical for training or evaluation.

Pure Rust Implementation

Zero-cost abstractions, safe concurrency, and native performance—ideal for large-scale simulations without garbage collection overhead.

Modular Game System

Cleanly separated components (skills, units, actions, rules) support rapid experimentation and adaptation to different game styles or rule sets.

Future Goals

The current goal is to complete the core simulator and use it in reinforcement learning experiments. Long-term, the project may evolve into an interactive tactical game prototype for demonstrating AI capabilities in complex decision environments. After completing simulator, we will apply Decision Transformer to this game.