English

LenGuard-GPC: Length Guarding with Guided-Prompt Consistency for Spatial Reasoning Reinforce Learning

Artificial Intelligence 2026-07-19 v1

Abstract

Multi-view spatial reasoning requires vision-language models to compare visual evidence across images, align object correspondences, and infer spatial relations over long visual contexts, a setting where chain-of-thought reasoning tends to grow verbose without becoming more accurate. Reinforcement learning with verifiable rewards is a natural fit for this task, but standard GRPO reward relies on sparse outcome-level feedback and gives no signal about where a reasoning trajectory goes wrong, nor any control over its length. We propose LenGuard-GPC, a dense reward framework that addresses both problems together. For each sampled trajectory, it compares the token-wise predictive distributions under a standard prompt and a guided prompt, and uses the resulting token-sum KL divergence as a dense reward signal. Since this KL penalty accumulates over tokens and would otherwise reward shorter responses regardless of their quality, we introduce a staged length bonus that keeps reasoning length within a controlled range without simply encouraging brevity. On six multi-view spatial reasoning benchmarks, LenGuard-GPC improves accuracy over vanilla GRPO while reducing average response length.

Keywords

Cite

@article{arxiv.2607.17243,
  title  = {LenGuard-GPC: Length Guarding with Guided-Prompt Consistency for Spatial Reasoning Reinforce Learning},
  author = {Xingjian Tao and Yiwei Wang and Yujun Cai and Jing Tang},
  journal= {arXiv preprint arXiv:2607.17243},
  year   = {2026}
}