CoSafe: Cooperative Runtime Safety Assurance for Mixed-Traffic Autonomous Driving

CoSafe: Cooperative Runtime Safety Assurance for Mixed-Traffic Autonomous Driving

Dingji Wang and coauthors argue that safety systems focused on one autonomous vehicle at a time can miss hazards created when several road users make individually reasonable but incompatible decisions. They introduce CoSafe, a framework that uses vehicle-to-everything (V2X) communication so roadside units can monitor hazards across a traffic scene and coordinate interventions involving multiple vehicles. The authors report integrating CoSafe with seven autonomous driving systems and evaluating it on the InterDrive benchmark. In that evaluation, they report higher driving scores, fewer safety violations, and accurate identification of unsafe vehicles while meeting real-time constraints. The paper’s central idea is to shift runtime safety assurance from an individual vehicle’s takeover to cooperative, scene-level intervention. The reported results are benchmark findings; the abstract does not establish how CoSafe would perform in real-world traffic.

Publication: ba63c38d-3cfa-40d9-81a0-378880f3a5a4

attribution — pass: The reading attributes the argument, framework and reported results to Dingji Wang and coauthors, matching the arXiv author list. Results are framed as what the authors report, not as established fact.

Authors:Dingji Wang, You Lu, Haowen Jiang, Yanggang Chen, Yuyang Bi, Bihuan Chen, Shenglin Bao, Xin Peng

summary — pass: The summary leads with the authors' argument, explains V2X and roadside units in plain language, and states the central shift from single-vehicle takeover to scene-level cooperative intervention. Minor drift: 'can miss hazards' and 'several road users' slightly paraphrase the source, which says ego-centric approaches cannot reliably resolve conflicts among multiple vehicles through independent takeover. A closer fix would be 'cannot reliably resolve' and 'multiple vehicles'. This does not change the meaning. The question is a fair open issue.

cannot be reliably resolved through independent takeover

date — pass: originalPublished 2026-11-04 matches the v1 submission date on arXiv. It is not the later feed date of 2026-11-05.

[Submitted on 4 Nov 2026]

support — pass: All three claims are directly supported by the abstract. The figures for driving score, safety violations and F3 score are quoted accurately. The limitation correctly confines the results to the InterDrive benchmark and does not turn them into real-world reliability. No unsupported benefits, costs or stakeholders are implied.

our evaluation on the InterDrive benchmark demonstrates that CoSafe improves the driving score of AVs by 70.03% on average, reduces safety violations by 87.80%

media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to access the source.