ProSR: Process-Shaped Spatial Reasoning for Reliable Chain-of-Thought in VLMs
For vision-language model developers, this work addresses process-level degradation in spatial reasoning, offering a method to enhance reliability beyond outcome-based training.
ProSR introduces a process-shaping optimization framework for spatial reasoning in VLMs, using Counterfactual Invariance and Tail Drift Penalties to improve visual dependence and trajectory stability, achieving higher accuracy on complex benchmarks.
Reliable spatial reasoning remains a core bottleneck for vision-language models (VLMs). Existing mainstream training paradigms for spatial reasoning largely rely on outcome alignment or process imitation, lacking explicit constraints on the reasoning process, and therefore struggle to ensure genuine visual dependence and stable reasoning trajectories. In this paper, we construct a high-quality CoT dataset covering diverse spatial phenomena and diagnose the model's reasoning process, revealing two typical types of process degradation during reinforcement learning optimization: Spurious Grounding, which bypasses visual evidence, and Tail Instability, where uncertainty abnormally rises in the later stage of reasoning. To address these issues, we propose ProSR, a process-shaping optimization framework for spatial reasoning. Through a Counterfactual Invariance Penalty and a Tail Drift Penalty, ProSR extends the optimization objective from single answer correctness to two process-level dimensions: visual dependence and trajectory stability. Experiments on multiple complex and out-of-distribution spatial reasoning benchmarks show that ProSR improves answer accuracy while generating reasoning trajectories that are more stable and more dependent on visual evidence.