arXiv cs.AI· Gautam Kumar Jain, Carsten Markgraf, Julian St\"ahler·· 3 天前AI 评分18
DriveLM-nuScenes 跨阶段一致性研究：显式提示与门控上下文路由对比
Cross-Stage Coherence in Hierarchical Driving VQA: A Consistency Protocol and Parameter-Efficient Gated Context Routing
AI 导读
研究在 DriveLM-nuScenes 上对比两种跨阶段上下文传递机制：显式变体在 4B 模型 Mini-InternVL2-4B-DA-DriveLM 上免训练使用三种提示策略。
正文 · 原文
Abstract:Graph Visual Question Answering (GVQA) for autonomous driving organizes reasoning into ordered stages, namely Perception, Prediction, and Planning, where planning decisions should remain consistent with the model's own perception. We present a comparative study of cross-stage context passing on DriveLM-nuScenes using two complementary mechanisms. The explicit variant evaluates three prompt-based conditioning strategies on a domain-adapted 4B VLM (Mini-InternVL2-4B-DA-DriveLM) without additional training, reducing NLI contradiction by up to 42.6% and establishing a strong zero-training baseline. The implicit variant introduces gated context projectors, which extract a hidden-state vector from one stage and inject a normalized, gated projection into the next stage's input embeddings. These projectors are jointly trained with stage-specific QLoRA adapters on a general-purpose 8B VLM (InternVL3-8B-Instruct) while updating only approximately 0.5% of parameters. The implicit variant achieves a statistically significant 34% reduction in planning-stage NLI contradiction (bootstrap 95% CIs, p < 0.05) and increases cross-stage entailment by 50%, evaluated with a multilingual NLI classifier to account for mixed-language outputs. Planning language quality also improves (CIDEr +30.3%), but lexical overlap and structural consistency degrade due to the absence of driving-domain pretraining. Since the two variants use different base models, we present them as complementary case studies: explicit context passing provides a strong training-free baseline for surface consistency, while implicit gated projection delivers significant planning-stage semantic gains, suggesting domain adaptation as a plausible next ingredient for full-spectrum improvement.
| Comments: | 17 pages, 8 figures, 8 tables. Accepted and presented at the COLM 2026 Workshop on Efficient Reasoning |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2604.22560 [cs.CV] |
| (or arXiv:2604.22560v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2604.22560 arXiv-issued DOI via DataCite |
Submission history
From: Gautam Kumar Jain [view email]
[v1]
Fri, 24 Apr 2026 13:54:48 UTC (14,243 KB)
[v2]
Tue, 13 Oct 2026 09:52:37 UTC (14,245 KB)
来源：arXiv cs.AI · arxiv.org