# CoSafe tests roadside coordination in simulation

> Researchers test roadside coordination in simulation.

By BIGCHANGE.AI Editorial

Published: 2026-11-05T15:32:38.557Z
Updated: 2026-11-05T15:32:38.557Z
Canonical: https://bigchange.ai/blog/cosafe-roadside-driving-study

![Conceptual illustration of cars at an intersection and a roadside communications unit.](https://bigchange.ai/api/media/file/cosafe-intersection-concept.png)
AI-generated conceptual illustration by BIG CHANGE.

A November 4 [preprint](https://arxiv.org/html/2611.06204v1) tests roadside coordination system CoSafe in simulation.

## The big change

- A roadside unit detects conflicts using observations and vehicle data.
- It selects coordinated paths for affected vehicles; their switches execute them while others retain their plans.
- Participating vehicles must share information and follow those paths; humans remain outside its control.

## Why coordination matters

At an unsignalized intersection, two conflicting cars may both wait; coordination lets one proceed while the other yields in the paper’s example.

The US Department of Transportation describes vehicle-to-everything communication as an ecosystem spanning vehicles, roadside equipment and supporting systems. Its [V2X briefing](https://www.itskrs.its.dot.gov/sites/default/files/2025-05/executive-briefing/Executive%20Briefing%202024_V2X%20Technology_FINAL508_05302025.pdf) identifies interoperability across manufacturers and jurisdictions as a challenge. Connected equipment must be able to communicate across those boundaries for information to be useful beyond a single setup.

## The experiment and its limits

Across seven driving systems and 46 routes, averaging five runs, the authors report 87.80% fewer violations across four categories versus unprotected baselines. Deadlocks and stalls remained.

For one system, delayed or lost messages weakened benefits; without a trajectory, vehicles used their own plans. The study excludes sensor and actuator faults and does not establish public-road performance.

The evaluation used [InterDrive](https://openaccess.thecvf.com/content/ICCV2025/html/Liu_CoLMDriver_LLM-based_Negotiation_Benefits_Cooperative_Autonomous_Driving_ICCV_2025_paper.html) with background traffic. Its earlier creators introduced the CARLA benchmark; they did not independently validate CoSafe.

## Sources

- [CoSafe preprint](https://arxiv.org/html/2611.06204v1) — Simulation methods and results.
- [InterDrive paper](https://openaccess.thecvf.com/content/ICCV2025/html/Liu_CoLMDriver_LLM-based_Negotiation_Benefits_Cooperative_Autonomous_Driving_ICCV_2025_paper.html) — Original benchmark research.
- [US DOT V2X briefing](https://www.itskrs.its.dot.gov/sites/default/files/2025-05/executive-briefing/Executive%20Briefing%202024_V2X%20Technology_FINAL508_05302025.pdf) — General V2X interoperability.