Mesh-RL: Coupled subgrid reinforcement learning

DOI10.48550/arXiv.2606.26333MaRDI QIDQ7270113arXivFDO

Authors Behnam Gheshlaghi, Bahador Rashidi, Shahin Atakishiyev

Publication date 24 June 2026

Published in arXiv

Tldr Mesh-RL accelerates temporal-difference learning by partitioning the environment into overlapping subgrids and enforcing boundary-consistent updates, improving convergence speed, cumulative reward, and stability in large or sparse-reward environments while preserving global coherence without modifying the reward function or introducing explicit planning mechanisms.

Full work available at URL https://doi.org/10.48550/arXiv.2606.26333

arXiv classification

cs.LG

This page was built for publication: Mesh-RL: Coupled subgrid reinforcement learning

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7270113)