Multi-Agent Coordination Using Large Language Models (LLMs)

Our work focuses on optimizing and utilizing large language models (LLMs) and Transformer architectures to improve multi-agent planning in embodied robotics and navigation. We examine the problem of effective coordination and planning in these environments by leveraging the rich latent space of Transformer architectures to facilitate multi-agent capabilities.

Related DINaMo Theses

• Learning Latent World Models for Multi-Agent Embodied Navigation and Planning, Richard Yun, EECS MEng ’26
• Stairway to Autonomy: Hierarchical Decision-Making for LLM-Guided Planning, Bandit-Driven Exploration, and Multi-Agent Navigation, Siddharth Nayak, AeroAstro PhD ’25
• Model-based Planning for Efficient Task Execution, Wenqi Ding, EECS MEng ’25
• Contextual Knowledge Sharing in Multi-Agent Long-Horizon Planning Settings with Centralized Communication and Coordination, Jackson Zhang, EECS MEng ’25

Publications

• Long-Horizon Planning for Multi-Agent Robots in Partially Observable Environments. (NeurIPS 2024)
• MAP-THOR: Benchmarking Long-Horizon Multi-Agent Planning Frameworks in Partially Observable Environments. (Multi-modal Foundation Model meets Embodied AI Workshop @ICML, 2024)

People

Siddharth Nayak, Jackson Zhang, Wenqi Ding, Richard Yun, Hamsa Balakrishnan