The Art of Interrogation: Consistency Amplifies Factuality in Spatial Reasoning

In three linesLarge Reasoning Models (LRMs) underperform on spatial reasoning tasks. Instead of supervised fine-tuning, authors propose a self-supervised RL framework using consistency verifiers (geometric and semantic transformations). They introduce OT-GRPO, an optimal transport-based variant of GRPO, achieving supervised-model performance without ground-truth labels.

Summary generated by Claude — human-verified