Conceitos Básicos
LLMs are utilized in SMART-LLM to efficiently generate multi-robot task plans by leveraging task decomposition, coalition formation, and skill-based task assignment.
Resumo
SMART-LLM introduces a framework for embodied multi-robot task planning using Large Language Models (LLMs). The process involves task decomposition, coalition formation, and task allocation guided by LLM prompts. A benchmark dataset is created for validation across various tasks. Evaluation experiments demonstrate promising results in simulation and real-world scenarios.
I. Introduction
Multi-robot systems enhance efficiency in various applications.
Skillful task allocation among heterogeneous robot arrays is crucial.
II. Related Works
Multi-robot task planning phases include decomposition, coalition formation, allocation, and execution.
III. Problem Formulation
Given high-level language instructions, the goal is to understand tasks and formulate executable plans for efficient robot utilization.
IV. Methodology
A. Stage 1: Task Decomposition
B. Stage 2: Coalition Formation
C. Stage 3: Task Allocation
D. Stage 4: Task Execution
V. Experiments
A. Benchmark Dataset includes elemental, simple, compound, and complex tasks.
B. Simulation Experiments evaluate SMART-LLM performance with different LLM backbones and baselines.
C. Real-Robot Experiments test visibility coverage tasks with varying regions and navigation challenges.
VI. Results and Discussion
Simulation experiments show SMART-LMM's adaptability across different categories of tasks.
Variability in performance observed across trials.
VII. Conclusions and Future Work
SMART-LMM demonstrates the potential of LLMs in generating multi-robot task plans efficiently with adaptability to new environments and scenarios.
Estatísticas
"Our method consistently achieves favorable outcomes irrespective of the LLM backbone employed."
"In compound and complex tasks, our method consistently achieves favorable results across all LLM backbones."