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

Jean Douglas Carvalho; Hugo Taciro Kenji; Ahmad Mohammad Saber; Glaucia Melo; Max Mauro Dias Santos; Deepa Kundur

cs.AIMar 29, 2026

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

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 presence, indicating that variability often reflects intrinsic semantic indeterminacy rather than isolated model failure. These findings highlight the importance of scenario-centric auditing and explicit ambiguity management when integrating LLM-based reasoning into safety-aligned driver assistance systems.

Figures & tables

Fig. 1 : Unified Multimodal Dataset: Data from heterogeneous vehicles are abstracted into a unified multimodal representation that jointly encodes visual perception, vehicle telemetry, and external contextual information within the multimodal data platform.

Fig. 2 : Deterministic scenario construction and consumption workflow. Normalized scene states $\mathcal{U}(a,t)$ , indexed at 1 Hz, are queried by backend data services to materialize scenario snapshots $\mathcal{S}(a,t)$ through the deterministic operator $\mathcal{B}(\cdot)$ under a structured query specification $\theta$ . Scenario construction combines semantic filters spanning visual perception, vehicle telemetry, weather context, and map information. The resulting scenarios are exposed consistently to both the web-based authoring interface for selection and inspection and the local testing module for scenario evaluation.

Fig. 3 : Scenario window generation from a single user-defined key moment ( $t_{0}$ ), illustrating controlled temporal expansion before and after the anchor instant.

Fig. 4 : Scenario-driven workflow of the platform, from web-based identification of key moments to local temporal expansion and structured prompt construction. Curated scenarios are transformed into bounded temporal windows and independently evaluated by multiple language models under identical conditions, thereby establishing a deterministic, reproducible path from interactive scenario selection to controlled model inference.

Fig. 5 : Mosaic of representative sixteen scene anchors, that serve as temporal seeds for subsequent scenario expansion, illustrating the diversity of near-people situations across urban layouts, traffic densities, lighting conditions, and pedestrian configurations.

Fig. 6 : Online visualization of the evaluated test sets available in the platform. Each entry corresponds to a near-people key-moment collection evaluated under identical conditions by different language models, including two text-only configurations and one multimodal model.

Fig. 7 : Mean overall risk level assigned by each evaluated language model across all scenario windows.

Fig. 8 : Percentage of scenario windows classified as high risk (risk level $\geq 4$ ) by each model.

Fig. 9 : Mean number of distinct evidence signals explicitly referenced in model outputs, reflecting reasoning breadth and justification externalization.

Fig. 10 : Distribution of dominant risk factors identified by each model. The six most frequent factors are shown explicitly, with the remaining factors grouped as Other.

Fig. 11 : Radar-based summary of inter-model behavior across near-people scenarios, highlighting systematic differences in risk attribution, high-risk escalation, VRU presence interpretation, evidence signal count (reasoning breadth), uncertainty expression, and relative token usage. The visualization illustrates how identical scenario artifacts yield divergent yet internally consistent interpretations across text-only and multimodal LLM configurations.

Fig. 12 : Compact heatmap of composite inter-model uncertainty across the evaluated scenarios. Rows correspond to scenario identifiers, ordered from lowest to highest uncertainty (bottom to top), and columns correspond to the evaluated language models. Color intensity indicates the per-model contribution to composite uncertainty, aggregated over overall risk severity, high-risk escalation, evidence usage, and dominant risk factor attribution.

LLM Auditing

Large language model auditing tests models and deployed systems for bias, safety, privacy, reliability, and provenance using behavioral and internal evidence.

See all 639 papers

DriveSafe: A Framework for Risk Detection and Safety Suggestions in Driving Scenarios
May 16, 2026Sainithin Artham, Shankar Gangisetty, Avijit Dasgupta +1VLMs for Autonomous Driving Autonomous Driving Safety Evaluation
IIIT-Hyderabad, India
UniDrive: A Unified Vision-Language and Grounding Framework for Interpretable Risk Understanding in Autonomous Driving
Jun 23, 2026Xiaowei Gao, Pengxiang Li, Yitai Cheng +4VLMs for Autonomous Driving Autonomous Driving Perception
Department of Earth Science & Engineering, Imperial College London, London, SW7 2AZ, United Kingdom · Department of Computing, The Hong Kong Polytechnic University, Hong Kong, China · SpaceTimeLab, Department of Civil, Environmental, and Geomatic Engineering, University College London, London, WC1E 6BT, United Kingdom +3
Reason--Imagine--Act: Closed-Loop LLM Decision Making with World Models for Autonomous Driving
May 19, 2026Zhengqi Sun, Yiwen Sun, Boxuan Liu +3Language Model-Based Control World Models
Department of Information Management, Peking University, Beijing 100871, China · School of Intelligence Science and Technology, Peking University, Beijing 100871, China · State Key Laboratory of General Artificial Intelligence, BIGAI, Beijing 100080, China +3

May 16, 2026J/K move · Enter open · S save

DriveSafe: A Framework for Risk Detection and Safety Suggestions in Driving Scenarios

Sainithin Artham, Shankar Gangisetty, Avijit Dasgupta, C. V. Jawahar

IIIT-Hyderabad, India

VLMs for Autonomous Driving Autonomous Driving Safety Evaluation Large Vision-Language Models Autonomous Driving Autonomous Driving Perception

Open paper AskPDF arXiv

Comprehensive situational awareness is essential for autonomous vehicles operating in safety-critical environments, as it enables the identification and mitigation of potential risks. Although recent Multimodal Large Language Models (MLLMs) have shown promise on general vision-language tasks, our findings indicate that zero-shot MLLMs still underperform compared to domain-specific methods in fine-grained, spatially grounded risk assessment. To address this gap, we propose DriveSafe, a framework for risk-aware scene understanding that leverages structured natural language descriptions. Specifically, our method first generates spatially grounded captions enriched with multimodal context, including motion, spatial, and depth cues. These captions are then used for downstream risk assessment, explicitly identifying hazardous objects, their locations, and the unsafe behaviors they imply, followed by actionable safety suggestions. To further improve performance, we employ caption-risk pairings to fine-tune a lightweight adapter module, efficiently injecting domain-specific knowledge into the base LLM. By conditioning risk assessment on explicit language-based scene representations, DriveSafe achieves significant gains over both zero-shot MLLMs and prior domain-specific baselines. Exhaustive experiments on the DRAMA benchmark demonstrate state-of-the-art performance, while ablation studies validate the effectiveness of our key design choices. Project page: https://cvit.iiit.ac.in/ research/projects/cvit-projects/drivesafe

Abstract

LLM Auditing

Explore similar work

DriveSafe: A Framework for Risk Detection and Safety Suggestions in Driving Scenarios

UniDrive: A Unified Vision-Language and Grounding Framework for Interpretable Risk Understanding in Autonomous Driving

Reason--Imagine--Act: Closed-Loop LLM Decision Making with World Models for Autonomous Driving