LinkedIn · 17 November 2025

LaGAT: neural search outperforms classic MAPF planners — AAAI 2026

Thrilled to share that our new paper has been accepted as an oral presentation at #AAAI2026! 🎉

We’ve been exploring how to truly hybridize learning and search for complex multi-agent planning — and we’re excited about what we found.

🔍 Graph Attention-Guided Search for Dense Multi-Agent Pathfinding Rishabh Jain Keisuke Okumura, Michael Amir & Amanda Prorok 📄 Paper: https://lnkd.in/e67nQsZW 👨‍💻 Code: https://lnkd.in/e8jrYBjq

In multi-agent pathfinding (#MAPF), learning-based methods have historically lagged behind strong search-based planners. So we asked: Can neural components really outperform SoTA search, and if so, how? For (arguably) the first time, our neural approach — LaGAT — clearly surpasses leading search-based planners 💥

The key: fusing GNN-based imitation learning with the LaCAM search planner to get the best of both worlds. Super excited for the discussions and energy at AAAI 2026 in Singapore 🇸🇬 this January!

More about the conference: https://lnkd.in/dvGu8Cqr

read the full post on LinkedIn →