aiqbench analysis
This research introduces the Shared Recurrent Memory Transformer (SRMT), a decentralized multi-agent reinforcement learning architecture that facilitates coordination through a global memory workspace without relying on centralized training. By enabling agents to broadcast and query working memory states, SRMT achieves emergent cooperation in complex pathfinding tasks even under sparse reward conditions, effectively scaling with larger agent populations and map sizes.
SRMT: Shared Memory for Multi-agent Lifelong Pathfinding
arXiv:2501.13200v2 Announce Type: replace-cross Abstract: Coordination in decentralized multi-agent reinforcement learning (MARL) necessitates that agents share information about their behavior and intentions. Existing approaches rely on communication protocols with domain or resource constraints or centralized training that poorly scales to large agent populations. We introduce the Shared Recurrent Memory Transformer (SRMT), which enables coordination through unconstrained communication. SRMT provides a global memory workspace where agents broadcast their learned working memory states and query others' memory representations to exchange information and coordinate while maintaining decentralized training and execution. We evaluate SRMT on the Partially Observable Multi-Agent Pathfinding (PO-MAPF) problem, where coordination is vital for optimal path planning and deadlock avoidance. We demonstrate that shared memory enables emergent coordination even when the reward function provides minimal or no guidance. On the specifically constructed Bottleneck task that requires negotiation, SRMT consistently outperforms communicative and memory-augmented baselines, particularly under sparse reward signals, and successfully generalizes to longer corridors unseen during training. On POGEMA maps, SRMT scales with the increasing agents' population and map size, achieving competitive performance with recent MARL, hybrid, and planning-based methods while requiring no domain-specific heuristics. These results demonstrate that a transformer with shared recurrent memory enhances coordination in decentralized multi-agent systems. The source code for training and evaluation is available on GitHub: https://github.com/Aloriosa/srmt.
Abstract:Coordination in decentralized multi-agent reinforcement learning (MARL) necessitates that agents share information about their behavior and intentions. Existing approaches rely on communication protocols with domain or resource constraints or centralized training that poorly scales to large agent populations. We introduce the Shared Recurrent Memory Transformer (SRMT), which enables coordination through unconstrained communication. SRMT provides a global memory workspace where agents broadcast their learned working memory states and query others' memory representations to exchange information and coordinate while maintaining decentralized training and execution. We evaluate SRMT on the Partially Observable Multi-Agent Pathfinding (PO-MAPF) problem, where coordination is vital for optimal path planning and deadlock avoidance. We demonstrate that shared memory enables emergent coordination even when the reward function provides minimal or no guidance. On the specifically constructed Bottleneck task that requires negotiation, SRMT consistently outperforms communicative and memory-augmented baselines, particularly under sparse reward signals, and successfully generalizes to longer corridors unseen during training. On POGEMA maps, SRMT scales with the increasing agents' population and map size, achieving competitive performance with recent MARL, hybrid, and planning-based methods while requiring no domain-specific heuristics. These results demonstrate that a transformer with shared recurrent memory enhances coordination in decentralized multi-agent systems. The source code for training and evaluation is available on GitHub: this https URL.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA)
arXiv:2501.13200 [cs.LG]
(or arXiv:2501.13200v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2501.13200 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 308 words
Related coverage
From theory to delivery: How Atos upskilled 400 engineers in agentic AI
AWS · Sep 1, 2026
MATES: Learning Multi-Agent Interactions by Transforming Observations for Frozen Single-Agent Policies
arXiv · Sep 24, 2026
AnyGoal: Vision-Language Guided Multi-Agent Exploration for Training-Free Lifelong Navigation
arXiv · Jun 16, 2026
ThinkFlow: Self-Evolving Probabilistic Latent Memory for Lifelong Conversational Agents
arXiv · Sep 16, 2026