Aug 2026· arXiv.org· Vol abs/2608.04905· 1 citation· 61 references
Computer Science
TL;DR
The resulting framework combines learned exploration with structured expert guidance without requiring LaCAM3 at inference and substantially outperforms state-of-the-art learning-based baselines and scales to ultra-large instances with up to city-level 100,000 agents.
Abstract
We present PRIMAL3, an ultra-large-scale learning-based framework for multi-agent pathfinding (MAPF) that integrates reinforcement learning, topology-aware communication, LaCAM3-guided training, and PIBT-based action refinement. PRIMAL3 targets failures at topologically critical states, where agents must coordinate decisively around bottlenecks, dead ends, and persistent conflicts. Each agent is represented using features derived from cut vertices, dead-end regions, shortest-path distances, and blocking estimates. Two complementary graphs capture agent interactions: a same-direction following graph propagates multihop context along compatible paths, while a different-direction conflict graph differentiates agents competing for shared space through masked attention and relative features. During training, we propose to let policy entropy identify uncertain agents, for which LaCAM3 provides confidence-triggered action interventions and label-smoothed imitation targets. During execution, a priority-aware PIBT module refines the proposed joint actions using persistent, learned, and distance-aware priorities together with policy-aware fallback preferences while maintaining collision-free execution. The resulting framework combines learned exploration with structured expert guidance without requiring LaCAM3 at inference. Experiments demonstrate that PRIMAL3 substantially outperforms state-of-the-art learning-based baselines and scales to ultra-large instances with up to city-level 100,000 agents. Real-world experiments further demonstrate the feasibility of deploying PRIMAL3 on physical robotic systems and ablation studies validate the individual contributions the components we proposed. Project page: https://marmotlab.github.io/PRIMAL3/
Multi-agent pathfinding (MAPF) requires multiple agents to reach assigned goals while avoiding vertex, edge, and intermittent-obstacle conflicts in a shared workspace. Learning-based MAPF is attractive for online decision-making, but fully decentralized reinforcement learning remains vulnerable to sparse rewards, non-s...
The results support a descriptive efficiency advantage for the integrated intelligent-control stack under the evaluated conditions, without establishing universal superiority or a formal safety guarantee.
EFXPLORER is proposed, a constrained exploration framework that maximizes exploration objectives subject to a constraint that preserves established task performance and incorporates a successor distance-based intrinsic reward that captures long-horizon dependencies.
Sungil Son, Hoseong Jung, Dahyun Oh et al.· 0 citations
In grid-world settings, LYCO consistently outperformed representative decentralized RL baselines in high-density environments, achieving a success rate above 0.7 at an agent density of 0.3 on a $20\times 20$ map with obstacle density 0.1 while maintaining a competitive makespan.
Hiroya Makino, Koji Noshiro, Koji Hasebe et al.· IEEE Access· 0 citations
Results show that round-level intent refinement can improve joint-action coordination while preserving decentralized execution, and DMM, the Decentralized Master-Mind, achieves higher success rates and lower solution costs than the evaluated learnable baselines.
Valeriy Vyaltsev, A. Andreychuk, T. Zlotnikova et al.· arXiv.org· 1 citation
This study explores a meta-level Mixture-of-Experts (MoE) approach to learning-based MAPF, where given a collection of compact yet high-performing experts specialised for different map distributions, the architecture learns a router that selectively fuses their outputs according to each agent's local situation.
Fumimaro Suzuki, Keisuke Okumura· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.