[2603.27536] Dual-Stage LLM Framework for Scenario-Centric Semantic Interpretation in Driving Assistance

Computer Science > Artificial Intelligence arXiv:2603.27536 (cs) [Submitted on 29 Mar 2026] Title:Dual-Stage LLM Framework for Scenario-Centric Semantic Interpretation in Driving Assistance Authors:Jean Douglas Carvalho, Hugo Taciro Kenji, Ahmad Mohammad Saber, Glaucia Melo, Max Mauro Dias Santos, Deepa Kundur View a PDF of the paper titled Dual-Stage LLM Framework for Scenario-Centric Semantic Interpretation in Driving Assistance, by Jean Douglas Carvalho and 5 other authors View PDF HTML (experimental) Abstract:Advanced Driver Assistance Systems (ADAS) increasingly rely on learning-based perception, yet safety-relevant failures often arise without component malfunction, driven instead by partial observability and semantic ambiguity in how risk is interpreted and communicated. This paper presents a scenario-centric framework for reproducible auditing of LLM-based risk reasoning in urban driving contexts. Deterministic, temporally bounded scenario windows are constructed from multimodal driving data and evaluated under fixed prompt constraints and a closed numeric risk schema, ensuring structured and comparable outputs across models. Experiments on a curated near-people scenario set compare two text-only models and one multimodal model under identical inputs and prompts. Results reveal systematic inter-model divergence in severity assignment, high-risk escalation, evidence use, and causal attribution. Disagreement extends to the interpretation of vulnerable road user prese...