LenGuard-GPC: Length Guarding with Guided-Prompt Consistency for Spatial Reasoning Reinforce Learning
Artificial Intelligence
Summary
The authors address the challenge of making AI better at reasoning about spatial relationships across multiple images without giving overly long or less accurate explanations. They develop LenGuard-GPC, a new method that provides more detailed feedback during learning by comparing predictions token by token. This approach also controls how long the AI's responses are, avoiding very short or excessively long answers. Their method improved accuracy while keeping answers concise on several spatial reasoning tests.
Authors
Xingjian Tao, Yiwei Wang, Yujun Cai, Jing Tang
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.