Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Augmented Runtime Collaboration for Self-Organizing Multi-Agent Systems: A Hybrid Bi-Criteria Routing Approach

Qingwen Yang, Feiyu Qu, Tiezheng Guo, Yanyi Liu, Yingyou Wen

TL;DR

This work tackles the challenge of task planning in open, decentralized multi-agent systems by introducing BiRouter, a runtime, local-information routing method. BiRouter uses a dual-criteria scoring scheme (ImpScore for long-term relevance and GapScore for contextual cohesion) combined with a dynamic agent-reputation mechanism to perform probabilistic successor selection, enabling emergent, globally coherent task pathways without a central planner. A large cross-domain dataset (MARS) was created to train the scoring functions, and BiRouter demonstrated superior performance, token efficiency, and robustness in both centralized and SO-MAS settings, including scenarios with unreliable agents. The approach promises scalable, adaptive collaboration for LLM-powered MAS in open environments and lays groundwork for online adaptive optimization and more complex collaboration patterns.

Abstract

LLM-based multi-agent systems have demonstrated significant capabilities across diverse domains. However, the task performance and efficiency are fundamentally constrained by their collaboration strategies. Prevailing approaches rely on static topologies and centralized global planning, a paradigm that limits their scalability and adaptability in open, decentralized networks. Effective collaboration planning in distributed systems using only local information thus remains a formidable challenge. To address this, we propose BiRouter, a novel dual-criteria routing method for Self-Organizing Multi-Agent Systems (SO-MAS). This method enables each agent to autonomously execute ``next-hop'' task routing at runtime, relying solely on local information. Its core decision-making mechanism is predicated on balancing two metrics: (1) the ImpScore, which evaluates a candidate agent's long-term importance to the overall goal, and (2) the GapScore, which assesses its contextual continuity for the current task state. Furthermore, we introduce a dynamically updated reputation mechanism to bolster system robustness in untrustworthy environments and have developed a large-scale, cross-domain dataset, comprising thousands of annotated task-routing paths, to enhance the model's generalization. Extensive experiments demonstrate that BiRouter achieves superior performance and token efficiency over existing baselines, while maintaining strong robustness and effectiveness in information-limited, decentralized, and untrustworthy settings.

Augmented Runtime Collaboration for Self-Organizing Multi-Agent Systems: A Hybrid Bi-Criteria Routing Approach

TL;DR

This work tackles the challenge of task planning in open, decentralized multi-agent systems by introducing BiRouter, a runtime, local-information routing method. BiRouter uses a dual-criteria scoring scheme (ImpScore for long-term relevance and GapScore for contextual cohesion) combined with a dynamic agent-reputation mechanism to perform probabilistic successor selection, enabling emergent, globally coherent task pathways without a central planner. A large cross-domain dataset (MARS) was created to train the scoring functions, and BiRouter demonstrated superior performance, token efficiency, and robustness in both centralized and SO-MAS settings, including scenarios with unreliable agents. The approach promises scalable, adaptive collaboration for LLM-powered MAS in open environments and lays groundwork for online adaptive optimization and more complex collaboration patterns.

Abstract

LLM-based multi-agent systems have demonstrated significant capabilities across diverse domains. However, the task performance and efficiency are fundamentally constrained by their collaboration strategies. Prevailing approaches rely on static topologies and centralized global planning, a paradigm that limits their scalability and adaptability in open, decentralized networks. Effective collaboration planning in distributed systems using only local information thus remains a formidable challenge. To address this, we propose BiRouter, a novel dual-criteria routing method for Self-Organizing Multi-Agent Systems (SO-MAS). This method enables each agent to autonomously execute ``next-hop'' task routing at runtime, relying solely on local information. Its core decision-making mechanism is predicated on balancing two metrics: (1) the ImpScore, which evaluates a candidate agent's long-term importance to the overall goal, and (2) the GapScore, which assesses its contextual continuity for the current task state. Furthermore, we introduce a dynamically updated reputation mechanism to bolster system robustness in untrustworthy environments and have developed a large-scale, cross-domain dataset, comprising thousands of annotated task-routing paths, to enhance the model's generalization. Extensive experiments demonstrate that BiRouter achieves superior performance and token efficiency over existing baselines, while maintaining strong robustness and effectiveness in information-limited, decentralized, and untrustworthy settings.

Paper Structure

This paper contains 32 sections, 11 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM-driven Multi-Agent Systems.
  4. Agent Collaboration Planning Mechanisms.
  5. Heuristic Search Methods.
  6. Methodology
  7. Preliminaries
  8. System and Agent Chain Definition.
  9. Problem Formulation.
  10. Forwarding Criteria
  11. Importance (ImpScore).
  12. Cohesion (GapScore).
  13. Agent Reputation.
  14. The Successor Selection Policy
  15. Probabilistic Synthesis.
  16. ...and 17 more sections

Figures (8)

  • Figure 1: The architecture of BiRouter and its workflow in self-organizing agent systems.
  • Figure 2: Distribution of MARS against others in a 2D semantic space projected by t-SNE.
  • Figure 3: BiRouter's performance and token consumption on GSM8K and MBPP across different $\alpha$ values.
  • Figure 4: Visualization of the performance metrics and token consumption. The diameter of each point is proportional to its $y$-axis value.
  • Figure 5: Agent credit score dynamics on GSM8K and HumanEval benchmarks.
  • ...and 3 more figures