Abstract:This study proposes LiP-LLM: integrating linear programming and dependency graph with large language models (LLMs) for multi-robot task planning. In order for multiple robots to perform tasks more efficiently, it is necessary to manage the precedence dependencies between tasks. Although multi-robot decentralized and centralized task planners using LLMs have been proposed, none of these studies focus on precedence dependencies from the perspective of task efficiency or leverage traditional optimization methods. It addresses key challenges in managing dependencies between skills and optimizing task allocation. LiP-LLM consists of three steps: skill list generation and dependency graph generation by LLMs, and task allocation using linear programming. The LLMs are utilized to generate a comprehensive list of skills and to construct a dependency graph that maps the relationships and sequential constraints among these skills. To ensure the feasibility and efficiency of skill execution, the skill list is generated by calculated likelihood, and linear programming is used to optimally allocate tasks to each robot. Experimental evaluations in simulated environments demonstrate that this method outperforms existing task planners, achieving higher success rates and efficiency in executing complex, multi-robot tasks. The results indicate the potential of combining LLMs with optimization techniques to enhance the capabilities of multi-robot systems in executing coordinated tasks accurately and efficiently. In an environment with two robots, a maximum success rate difference of 0.82 is observed in the language instruction group with a change in the object name.

What problem does this paper attempt to address?

### Problems the paper attempts to solve The paper "LiP - LLM: Integrating Linear Programming and dependency graph with Large Language Models for multi - robot task planning" aims to solve several key challenges in multi - robot task planning: 1. **Task - dependency management**: In a multi - robot system, there are precedence - dependency relationships between tasks (i.e., certain tasks can only be executed after other tasks are completed). These dependencies are crucial for the efficient execution of tasks. Although existing multi - robot task - planning methods use large language models (LLMs), they rarely focus on task efficiency or use traditional optimization methods to manage these dependencies. 2. **Task - assignment optimization**: In order to make multiple robots execute tasks more efficiently, task assignment needs to be optimized. This not only involves the sequential arrangement of tasks but also needs to consider the characteristics and capabilities of each robot to ensure the feasibility and efficiency of tasks. 3. **Natural - language - instruction processing**: A key challenge in multi - robot task planning is to extract tasks from natural - language instructions and ensure that these tasks are executed in the correct order. In addition, tasks without dependencies can be executed in parallel, further improving efficiency. ### Solutions The paper proposes a new multi - robot task - planning method - LiP - LLM, which combines linear programming (LP) and dependency graph with large language models (LLMs). The specific steps are as follows: 1. **Skill - list generation**: Use LLMs to generate a series of skills required to complete the given language instructions. By calculating the probability of each skill, ensure that the generated skills are feasible. 2. **Dependency - graph generation**: Based on the generated skill list, use LLMs to construct a dependency graph, which represents the dependency relationships between each skill. This step ensures the correct order of tasks. 3. **Task assignment**: Use the linear programming method to optimally assign tasks according to the dependency graph and the characteristics of each robot. By removing completed nodes, tasks are executed step by step. ### Experimental results The experiments were carried out in a simulated environment to evaluate the performance of LiP - LLM in different environments. The results show that, compared with existing multi - robot task - planning methods, LiP - LLM performs better in terms of success rate and efficiency. Specifically: - **Success rate**: LiP - LLM achieved the highest success rate in all experimental environments, with a maximum success rate of 0.82. - **Path - length - weighted success rate (SPL)**: LiP - LLM also performs well on the SPL index, indicating that the plans it generates are more efficient. - **Processing time**: Although the processing time is relatively long, the processing time of LiP - LLM is still better than other methods. ### Conclusion The paper demonstrates the potential of the method of combining linear programming and dependency graph with large language models in multi - robot task planning. This method can not only effectively manage task - dependency relationships but also optimize task assignment and improve the overall performance of multi - robot systems.

LiP-LLM: Integrating Linear Programming and dependency graph with Large Language Models for multi-robot task planning

MLDT: Multi-Level Decomposition for Complex Long-Horizon Robotic Task Planning with Open-Source Large Language Model

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

DELTA: Decomposed Efficient Long-Term Robot Task Planning using Large Language Models

SELP: Generating Safe and Efficient Task Plans for Robot Agents with Large Language Models

DART-LLM: Dependency-Aware Multi-Robot Task Decomposition and Execution using Large Language Models

SMART-LLM: Smart Multi-Agent Robot Task Planning using Large Language Models

LLM+P: Empowering Large Language Models with Optimal Planning Proficiency

Leveraging LLMs, Graphs and Object Hierarchies for Task Planning in Large-Scale Environments

Enhancing Robot Task Planning and Execution through Multi-Layer Large Language Models

ISR-LLM: Iterative Self-Refined Large Language Model for Long-Horizon Sequential Task Planning

Nl2Hltl2Plan: Scaling Up Natural Language Understanding for Multi-Robots Through Hierarchical Temporal Logic Task Representation

Task Planning for Robot Manipulator Using Natural Language Task Input with Large Language Models

A Framework for Neurosymbolic Robot Action Planning using Large Language Models

Large Language Models for Robotics: Opportunities, Challenges, and Perspectives

COHERENT: Collaboration of Heterogeneous Multi-Robot System with Large Language Models

MHRC: Closed-loop Decentralized Multi-Heterogeneous Robot Collaboration with Large Language Models

Large Language Models for Orchestrating Bimanual Robots