Orchestrator: Active Inference for Multi-Agent Systems in Long-Horizon Tasks

2025/09/06 by Beckenbauer, Lukas, Loewe, Johannes-Lucas, Zheng, Ge +1 · 2 citations
#Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Multiagent Systems (cs.MA)

paper · doi:10.48550/arxiv.2509.05651

Abstract

Complex, non-linear tasks challenge LLM-enhanced multi-agent systems (MAS) due to partial observability and suboptimal coordination. We propose Orchestrator, a novel MAS framework that leverages attention-inspired self-emergent coordination and reflective benchmarking to optimize global task performance. Orchestrator introduces a monitoring mechanism to track agent-environment dynamics, using active inference benchmarks to optimize system behavior. By tracking agent-to-agent and agent-to-environment interaction, Orchestrator mitigates the effects of partial observability and enables agents to approximate global task solutions more efficiently. We evaluate the framework on a series of maze puzzles of increasing complexity, demonstrating its effectiveness in enhancing coordination and performance in dynamic, non-linear environments with long-horizon objectives.

Citations

• MazeEval: A Benchmark for Testing Sequential Decision-Making in Language Models
• Optimizing Sequential Multi-Step Tasks with Parallel LLM Agents
• The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
• Adaptive Graph Pruning for Multi-Agent Communication
• Multi-Agent Collaboration via Evolving Orchestration
• Deep Active Inference Agents for Delayed and Long-Horizon Environments
• MAS-ZERO: Designing Multi-Agent Systems with Zero Supervision
• Cooperative Hybrid Multi-Agent Pathfinding Based on Shared Exploration Maps
• SagaLLM: Context Management, Validation, and Transaction Guarantees for Multi-Agent LLM Planning
• Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks
• AlphaMaze: Enhancing Large Language Models' Spatial Intelligence via GRPO
• Free Energy Risk Metrics for Systemically Safe AI: Gatekeeping Multi-Agent Study
• Teaching Language Models to Critique via Reinforcement Learning
• TradingAgents: Multi-Agents LLM Financial Trading Framework
• AMaze: An intuitive benchmark generator for fast prototyping of generalizable agents
• Factorised Active Inference for Strategic Multi-Agent Interactions
• AdaSociety: An Adaptive Environment with Social Structures for Multi-Agent Decision-Making
• Optima: Optimizing Effectiveness and Efficiency for LLM-Based Multi-Agent System
• Grounded Answers for Multi-agent Decision-making Problem through Generative World Model
• HDFlow: Enhancing LLM Complex Problem-Solving with Hybrid Thinking and Dynamic Workflows
• EVINCE: Optimizing Multi-LLM Dialogues Using Conditional Statistics and Information Theory
• LLaMAR: Long-Horizon Planning for Multi-Agent Robots in Partially Observable Environments
• Maximizing utility in multi-agent environments by anticipating the behavior of other learners
• AI Agents That Matter
• Adaptive and Optimal Second-order Optimistic Methods for Minimax Optimization
• Distributed maze exploration using multiple agents and optimal goal assignment
• LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision and the Road Ahead
• LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision, and the Road Ahead
• More Agents Is All You Need
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• On Implementing Autonomous Supply Chains: a Multi-Agent System Approach
• Memory Gym: Towards Endless Tasks to Benchmark Memory Capabilities of Agents
• AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors
• Self-Refine: Iterative Refinement with Self-Feedback
• Reflexion: Language Agents with Verbal Reinforcement Learning
• Multi-Agent Coordination via Multi-Level Communication
• Training Verifiers to Solve Math Word Problems
• Multi-Agent Pathfinding: Definitions, Variants, and Benchmarks
• Actor-Attention-Critic for Multi-Agent Reinforcement Learning
• HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering

Cited by

• AnimAgents: Coordinating Multi-Stage Animation Pre-Production with Human-Multi-Agent Collaboration
• What the flock knows that the birds do not: exploring the emergence of joint agency in multi-agent active inference

Related

• Learning Latency-Aware Orchestration for Multi-Agent Systems
• Multi-Agent Collaboration via Evolving Orchestration
• Maestro: Learning to Collaborate via Conditional Listwise Policy Optimization for Multi-Agent LLMs
• Neural Orchestration for Multi-Agent Systems: A Deep Learning Framework for Optimal Agent Selection in Multi-Domain Task Environments
• When Should We Orchestrate Multiple Agents?