LaMMA-P: Generalizable Multi-Agent Long-Horizon Task Allocation and Planning with LM-Driven PDDL Planner

  • 2024-09-30 18:58:18
  • Xiaopan Zhang, Hao Qin, Fuquan Wang, Yue Dong, Jiachen Li
  • 0

Abstract

Language models (LMs) possess a strong capability to comprehend naturallanguage, making them effective in translating human instructions into detailedplans for simple robot tasks. Nevertheless, it remains a significant challengeto handle long-horizon tasks, especially in subtask identification andallocation for cooperative heterogeneous robot teams. To address this issue, wepropose a Language Model-Driven Multi-Agent PDDL Planner (LaMMA-P), a novelmulti-agent task planning framework that achieves state-of-the-art performanceon long-horizon tasks. LaMMA-P integrates the strengths of the LMs' reasoningcapability and the traditional heuristic search planner to achieve a highsuccess rate and efficiency while demonstrating strong generalization acrosstasks. Additionally, we create MAT-THOR, a comprehensive benchmark thatfeatures household tasks with two different levels of complexity based on theAI2-THOR environment. The experimental results demonstrate that LaMMA-Pachieves a 105% higher success rate and 36% higher efficiency than existingLM-based multi-agent planners. The experimental videos, code, and datasets ofthis work as well as the detailed prompts used in each module are available athttps://lamma-p.github.io.

 

Quick Read (beta)

loading the full paper ...