A Google Research Football Multi-Agent Reinforcement Learning Library (GRF MARLLib)

11-vs-11 full-game replay

GRF-MARLLib Handbook

• Introduction
• Installation
• Scenarios
  • GRF intro
  • Scenario Description
• Framework Architecture
  • Workflow
  • Components
• Quick Start Benchmarking
  • Execution
  • Configuration file
  • Monitoring & Saving

Algorithm Documentation

• Cooperative Learning — MARL
  • Independent Learning
  • Centralized Training Decentralized Execution (CTDE)
  • Fully Decentralized Learning
• Competitive Learning — PBT
  • Policy-Space Response Oracle (PSRO)

ToolKits

• Pre-trained Policies
  • Potential Uses of Pre-trained GRF models
  • 5-vs-5 Radar Plots
  • 11-vs-11 Radar Plots
• Single-Step Visual Debugger
• Online Ranking
  • JIDI Intro
  • Daily Evaluation
  • Competition Evaluation
  • Tutorial