Semi-Automated Synthesis of Driving Rules
DOI: 10.14722/vehiclesec.2023.23083
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 challenge of encoding societal "common sense" driving rules for autonomous vehicles (AVs) to monitor for deviations that may indicate system failures or attacks. While existing AV security research focuses primarily on safety, the authors argue that AVs must also adhere to location-specific norms and best practices documented in state driving manuals. The study presents a semi-automated framework to extract these rules from natural language text, encode them in a formal specification, and make them available for anomaly detection and behavioral monitoring. The methodology involves collecting driving manuals from 15 U.S. states and six Australian territories. The authors filtered the PDFs to focus on safety-related sections, excluding administrative content. They developed a natural language processing pipeline that identifies sentences containing keywords such as "IF," "THEN," or "BECAUSE." These sentences are parsed into antecedent-consequent pairs and converted into RDF-style triples using part-of-speech tagging. The resulting rules are represented in an abstract, portable format: "IF(?x), THEN(?y)." The system automatically extracted 67 rules, which were then manually refined alongside 205 additional rules to create a corpus of 278 valid driving rules. The authors also performed meta-analyses to cluster rules by features such as equipment dependency or regional specificity. The results indicate significant variation in rule extraction success across regions, largely due to differences in manual formatting and writing styles. California yielded the highest number of rules (38 total), as its manual frequently used explicit if-then statements. Conversely, states like Tasmania and West Australia yielded few or no rules because their manuals relied heavily on diagrams, images, or non-standard listing formats that the parser could not process. The authors identified three primary sources of error: PDF parsing failures resulting in broken text, reliance on visual diagrams rather than text, and writing styles that did not match the expected conditional syntax. The final dataset includes rules ranging from basic safety instructions to complex interactions with law enforcement and environmental conditions. The significance of this work lies in providing the first publicly available corpus of formalized, common-sense driving rules derived from official handbooks. By encoding these rules in a standardized, interpretable format, the authors enable the development of anomaly detection systems that can verify whether an AV’s behavior aligns with expected societal norms. This approach complements existing formal logic and interlock systems by offering a high-level, abstract representation of driving expertise that is adaptable across different software environments. The authors conclude that this framework facilitates the monitoring of AVs for unexplained anomalies, distinguishing between legitimate contextual adjustments (e.g., slowing down for a dust storm) and potential security threats or system failures.
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | Crossref | — | — | 1 | 2026-08-09 |
| archive | success | canonical_url | — | — | 1 | 2026-08-09 |
| extract | success | cached | — | — | 3 | 2026-08-10 |
| clean | success | clean | — | — | 1 | 2026-08-09 |
| chunk | success | chunk | — | — | 1 | 2026-08-09 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-08-09 |
| promote | success | — | — | — | 1 | 2026-08-09 |
| summarize | success | llm | qwen3.6-27b-nvidia | summ-v5 | 2 | 2026-08-10 |
| tag | success | vector_similarity | — | — | 10 | 2026-08-11 |
| verify | success | — | — | — | 1 | 2026-08-10 |
Summary generated by qwen3.6-27b-nvidia on 2026-08-10; 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).
- Methodological Resource: dataset resource
- Theoretical Contribution: computational model, conceptual framework