RiskNet: interaction-aware risk forecasting for autonomous driving in long-tail scenarios
DOI: 10.1016/j.tre.2025.104478
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the critical challenge of ensuring autonomous vehicle (AV) safety in "long-tail" scenarios—rare but high-stakes driving conditions characterized by complex multi-agent interactions and high behavioral uncertainty. Traditional risk assessment metrics, such as Time to Collision (TTC) and Responsibility-Sensitive Safety (RSS), rely on deterministic assumptions that fail to capture the stochasticity of real-world traffic. Motivated by statistical analysis of fatal crash data, which reveals that risk arises from the coupling of infrastructure, vehicle types, and collision modes, the authors propose RiskNet. This framework aims to provide a unified, interaction-aware risk forecasting system that integrates physics-based modeling with data-driven probabilistic prediction. RiskNet employs a two-part methodology. First, it utilizes a deterministic field-theoretic model to quantify risk as an interaction field generated by the ego vehicle, surrounding agents, and infrastructure. This model calculates interaction energy and force based on physical properties (mass, velocity) and environmental constraints. To address directional sensitivity, the model incorporates a Doppler-effect-inspired adjustment that amplifies risk perception for approaching agents and attenuates lateral threats. Second, to handle behavioral uncertainty, the framework integrates a Graph Neural Network (GNN) combined with Neural Ordinary Differential Equations. This module predicts multi-modal future trajectory distributions for surrounding agents, capturing discrete social dependencies and continuous motion evolution. These probabilistic predictions are fused with the deterministic risk field to enable dynamic, time-horizon risk inference. The framework was evaluated on the highD, inD, and rounD datasets, covering diverse scenarios such as highway lane changes, intersection turns, and roundabout merges. Results demonstrate that RiskNet significantly outperforms traditional approaches (TTC, THW, RSS, and NC Field) in terms of accuracy, responsiveness, and directional sensitivity. The model exhibits strong generalization across different driving environments and maintains robust performance in high-risk, long-tail settings. By effectively mapping probabilistic behavior predictions into structured risk measures, RiskNet enables proactive safety assessment under uncertainty. The significance of this work lies in its ability to bridge the gap between deep learning-based trajectory prediction and interpretable, operational risk assessment. By combining physics-inspired interaction modeling with learned behavioral uncertainty, RiskNet offers a scalable foundation for safety-critical decision-making in autonomous driving. It supports real-time, scenario-adaptive risk forecasting, addressing the limitations of existing methods in unstructured or highly interactive urban settings. This approach enhances AV reliability in complex environments where traditional deterministic metrics are insufficient.
Key finding
RiskNet significantly outperforms traditional risk assessment metrics in accuracy, responsiveness, and directional sensitivity by integrating deterministic interaction field modeling with probabilistic trajectory prediction.
Methodology
simulation_modeling
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 1 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-07 |
| chunk | success | chunk | — | — | 1 | 2026-06-07 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-07 |
| enrich | failed | — | — | — | 5 | 2026-07-02 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model