RL for Provably Optimal Quadrilateral Block Decompositions
September 25, 2026
An RL agent is trained to perform local edits on half-edge data structures to reach the theoretical floor of vertex irregularity in planar meshes. The policy uses mesh-conforming convolutions, allowing it to generalize to larger, unseen domains.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates the utility of RL in solving topologically constrained geometric optimization problems.