Esquilax

JAX multi-agent system simulation toolset

Esquilax is a set of utilities and transformations implementing common patterns seen in multi-agent systems, allowing developers and researchers to quickly build models without the need to re-implement or optimise low level algorithms.

Esquilax is mainly intended for multi-agent RL, neuro-evolution and alife use-cases, and can be used alongside other JAX libraries like Flax, Evosax, and RLax.

Features

• Built on top of JAX, allowing for high performance and GPU support from JIT compiled Python.

• Interoperability with existing JAX ML, RL, and neuro-evolution libraries. Also works alongside the broader Python scientific/numerical ecosystem.

• Built-in multi-agent RL and neuro-evolution functionality. Use Esquilax simulations as multi-agent training environments.

• Functional paradigm allows models to be easily combined and re-used.

• Performant implementations of common multi-agent patterns.

Contents:

• Getting Started
  • Installation
• User Guide
  • Concepts
  • Tips
• Sharp-Bits
  • JAX Constraints
  • Unsupported Features
• Examples
  • Hard-Coded Boids
  • Evolutionary Boids
  • RL Boids
• API Reference
  • esquilax