ESCHER-Swarm-Solver is a comprehensive framework for solving and analyzing large-scale Swarm Air Defense scenarios using the ESCHER (Efficient Sampling of Checkpoints for History-based Equilibrium Recovery) algorithm.

This repository contains the game environments (built on OpenSpiel), the solver implementations (PyTorch), and visualization tools for analyzing strategies in both abstract and high-fidelity continuous environments.

A continuous, non-abstracted simulation with 40 attackers and 20 interceptors.

Visualizing the decision-making process in the discrete OpenSpiel environment.

Static analysis of engagement outcomes, showing drone paths, intercept points, and coverage.

We have detailed internal documentation explaining the theoretical foundations of the solvers used in this project:

• Guide to ESCHER: An in-depth explanation of the ESCHER algorithm, its advantages over MCCFR, and how it handles large imperfect-information games.
• Guide to CFR: A primer on Counterfactual Regret Minimization, the foundational algorithm for solving extensive-form games.

The project is organized into several modular components:

• ESCHER-Torch: A PyTorch reimplementation of the ESCHER algorithm. It includes the core solver logic and scripts to run it on standard OpenSpiel games (like Leduc Poker) and our custom Swarm environments.

• Swarm-AD-OpenSpiel: The core sequential, two-player OpenSpiel environment. It models a 10-drone swarm vs. a layered air defense system (autonomous launchers + interceptors) on a discrete grid.
• Swarm-AD-Large-OpenSpiel: A high-fidelity variant designed to stress-test strategies. It features 40 attackers, continuous space movement, directional AD behavior, and RRT-based pathfinding.

• Swarm-AD-Naive-Strategy: A simplified runner that wires blueprint strategies directly into the OpenSpiel game for quick visualization and debugging.
• Visualizer: A standalone package for generating Cursor-on-Target (CoT) events to visualize game states in WinTAK. It also handles the generation of the GIFs shown above.
• WinTAK-CoT-Generator: Utilities for generating CoT sequences for integration with tactical situational awareness tools.

To run the solver on the standard Swarm Defense game:

cd ESCHER-Torch
python run_escher_torch_swarm.py

To generate a visualization of a blueprint strategy:

cd Swarm-AD-Naive-Strategy
python run_blueprint_demo.py

Check the output/ directory for the resulting PNG.

To run the high-fidelity simulation and generate an animation:

cd Swarm-AD-Large-OpenSpiel
python demo_animation.py