LAMARL Framework

This paper introduces LAMARL, a novel approach that integrates Multi-Agent Reinforcement Learning (MARL) with Large Language Models (LLMs) to enhance sample efficiency and automate function generation for multi-robot cooperative tasks.

• 🎯 185.9% improvement in sample efficiency on average
• 🤖 Fully automated prior policy and reward function generation
• 🔧 28.5%-67.5% improvement in LLM output success rates through structured prompting
• ✅ Validated on both simulation and real-world shape assembly tasks