Open-source tools for Differential Games and Multi-Agent Control in Julia
The JuliaDifferentialGames organization builds a high-performance, research-grade ecosystem for modeling, solving, and learning differential games and multi-agent dynamical systems. Our libraries integrate modern numerical optimal control, dynamic programming, kinodynamic motion planning, and inverse problem methods to support advanced research and real-world applications.
This organization is under active development, with several core packages already available and additional libraries in progress.
⚠️ Work in progressThese libraries are currently under active development and not all capabilites have been released yet, but will (hopefully) be soon.
Core library for differential game modeling and solution methods.
Key capabilities:
- Continuous and discrete-time differential games
- Zero-sum, general-sum, and cooperative formulations
- Support for Multi-agent Reinforcement Learning (MARL) algorithms
- Open-loop and feedback Nash equilibria
- Customizable dynamics, costs, and constraints
- Interoperable with Julia’s scientific computing ecosystem
This package serves as the foundation for multi-agent optimal control and game-theoretic analysis in Julia.
Kinodynamic motion planning for warm-starting differential games.
Designed for dynamic and high-dimensional systems:
- Kinodynamic planning (RRT, RRT*, variants)
- Planning-to-optimization workflows
- Warm-start generation for multi-agent game solvers
- Tools for global strategy initialization
Ideal for initializing complex trajectories before game-theoretic refinement.
Solvers for nonlinear inverse problems
Provides advanced iterative methods:
- Landweber Method
- Iteratively Regularized Gauss–Newton (IRGN)
- Nonlinear multigrid regularization
This package is a general package for numerical analysis of inverse problems.
Learning-based optimization and amortized solvers.
Planned features:
- Neural warm-starts for repeated game solves
- Differentiable optimization layers
- Meta-learning for value functions and costates
- Integration with Flux, Zygote, and scientific machine learning ecosystems
Targeted toward differential games and optimal control.
Control Barrier Functions integrated directly into optimization problems.
Focus areas:
- Optimization-native CBF formulations
- Multi-agent safety constraints embedded in game objectives
- Scalable feasibility and safety enforcement
- Differentiable CBF integration for closed-loop control
This package will support rigorous safety-critical control in multi-agent settings.
JuliaDifferentialGames aims to provide a unified, composable environment for:
- Differential games and multi-agent optimal control
- Dynamic programming and motion planning
- Inverse differential games
- Learning-enhanced optimization methods
- Safety-critical control and control barrier function design
By building on Julia’s strengths in scientific computing, we seek to deliver tools that are fast, expressive, and suitable for cutting-edge research in robotics, aerospace, autonomous systems, and computational game theory.
We welcome contributions from the community. Early involvement is especially valuable as interfaces evolve.
Ways to contribute:
- Propose features or improvements through GitHub Issues
- Participate in API design discussions
- Contribute solvers, benchmarks, or documentation
- Share example notebooks or case studies
A full CONTRIBUTING guide will be added as the ecosystem matures.
Packages within this organization are generally released under the MIT License unless otherwise specified.
The ecosystem is actively expanding. APIs and interfaces may change as new solvers and workflows are integrated. For research or production use, we recommend pinning versions and tracking release notes.
- a visual diagram of the ecosystem
- a project logo in SVG format