DynaHMRC: Decentralized Heterogeneous Multi-Robot Collaboration for Dynamic Tasks with Large Language Models

2026-06-12

AI Summary

This work addresses key limitations of centralized large language model (LLM) scheduling in multi-robot systems—namely poor scalability, weak adaptability to dynamic tasks, and inefficient domain-specific reasoning due to scarce training data. To overcome these challenges, the authors propose a decentralized collaborative framework wherein heterogeneous robots operate as role-aware LLM agents. These agents coordinate through a four-phase closed-loop mechanism comprising self-description, leadership bidding, leader election, and reflective execution to accomplish dynamic tasks. The approach introduces the first decentralized architecture for heterogeneous multi-robot systems tailored to dynamic tasking, circumventing single-model context constraints while enabling flexible team formation and efficient coordination. A comprehensive benchmark encompassing diverse task types and dynamic environments is also established. Experiments demonstrate that the proposed method significantly improves task success rates, reduces both action and communication overhead, and exhibits strong scalability with increasing team size compared to strong baselines.

Abstract

Large language models (LLMs) provide robots with richer task understanding and adaptability, making them promising for coordinating heterogeneous multi-robot systems in long-horizon tasks. Despite this potential, several challenges remain underexplored: (1) Centralized LLM schedulers scale poorly as team size and environmental complexity increase. A single model must process excessive contextual information, and long-context approximation may degrade reasoning quality; (2) Existing task formulations insufficiently consider dynamic settings, while robust adaptation to evolving task conditions is essential for real-world deployment; (3) Domain-specific data scarcity limits specialized robotic reasoning, making proprietary general-purpose models inefficient for expert tasks. To address these limitations, we propose DynaHMRC, a decentralized framework in which each robot acts as a role-aware LLM agent. This design mitigates the single-model context bottleneck and supports flexible collaboration across heterogeneous team configurations. DynaHMRC organizes collaboration as a four-stage closed-loop process: self-description, task allocation with leadership bidding, leader election, and reflective execution, supported by executable robot interfaces. We further develop a benchmark covering three task families, four dynamic variations, and six team configurations to systematically study dynamic task modeling. In addition, we conduct an empirical analysis to guide the construction of domain-specific expert datasets and fine-tune pretrained LLMs to improve specialized competence. Experiments show that DynaHMRC achieves higher success rates than strong baselines with fewer action and communication steps, while demonstrating promising scalability trends as team size grows within the evaluated settings.

Problem

• decentralized coordination
• heterogeneous multi-robot systems
• dynamic tasks
• large language models
• domain-specific reasoning

Innovation

• Decentralized Multi-Robot Collaboration
• Large Language Models
• Dynamic Task Adaptation
• Heterogeneous Robot Teams
• Role-Aware LLM Agents