Safe deployment of autonomous vehicles in Urban environments

Background

Automated driving systems are advancing rapidly, but the capacity of cities and mobility authorities to consistently evaluate, approve, and monitor deployments has not scaled at the same pace. Each deployment is area-specific and must account for infrastructure conditions, traffic behaviour, environmental variability, ethical considerations, and public acceptance. As operational variations increase, authorities face growing approval backlogs, inconsistent decisions across comparable deployments, and heightened public scrutiny following incidents. In complex deployments, this sometimes manifests as “decision paralysis”. Traditional document-based approval processes are not designed to manage this level of multidimensional variability. Safe introduction, therefore, requires a structured and repeatable governance cycle in which deployment is incremental, evidence-based, and continuously monitored rather than treated as a one-time authorisation event.

Besides existing standards such as SOTIF and UL 4600, recently (Nov 2025), UNECE released a new regulation on automated driving systems, ECE-TRANS-WP.29-GRVA-2026-02, which emphasises structured safety assurance, traceable Operational Design Domain (ODD) definitions, scenario-based evaluation, performance monitoring, and harmonised approaches across contracting parties.

The Resembler platform supports these initiatives by using a contextual AI engine combined with hundreds of millions of kilometres of collected incidents and accidents. Resembler is a digital planning and deployment governance platform that supports cities and mobility authorities in the safe, incremental rollout of automated driving and robotic mobility systems.

Structured and Safe Deployment

The limiting factor for deployment of autonomous systems is no longer engineering feasibility - it is the ability of public authorities to consistently evaluate, approve, and monitor deployments across real, heterogeneous urban environments. Each deployment varies by area, infrastructure, traffic behaviour, environmental conditions, vehicle types, and operating constraints. This creates area-specific risk variability and a rapidly growing number of operational permutations that are difficult to compare and govern using traditional, document-based processes.

Regulatory approaches to automated driving also differ significantly across regions. Some jurisdictions rely on manufacturer self-certification and local permit frameworks, while others operate under structured type-approval regimes and detailed certification guidelines, such as those developed under UN-ECE frameworks. This diversity creates additional complexity for cities and operators seeking consistency, traceability, and comparability across deployments. Resembler provides a harmonizing digital layer that structures evidence, verification plans, monitored performance data, and ODD definitions in a format that can align with both self-approval environments and formal certification processes. The platform does not replace regulatory authority; rather, it supports it by translating deployment evidence into a consistent, auditable framework adaptable to different regional approval models.

As a result, cities face predictable outcomes: slow approvals, inconsistent decisions between comparable deployments, and limited transparency on why a deployment was accepted or rejected until incidents trigger public backlash. The challenge is not only technical and regulatory; it is also ethical and societal. Similar to other high-impact technologies introduced into public life, automated driving must be deployed in incremental steps, with clear evidence, continuous tracking, and the ability to detect and correct risk early.

Resembler addresses this governance gap by providing a platform that structures deployment decisions around the local context of a defined area of operation. It ingests the area of deployment, captures environmental and infrastructure constraints, and links these to a database of real-world accidents and incidents. This enables cities to build, maintain, and update a safety case for specific ODDs and vehicle types, and to continuously validate deployment changes through A/B testing against real-world evidence. Resembler supports the following key features

• Area-specific ODD definitions and constraints,

• Scenario and accident archetype mapping,

• Verification test plan generation,

• Structured evidence linking real-world accident data to deployment conditions,

• Monitored performance tracking over time, and

• Auditable documentation of incremental deployment decisions.

By translating regulatory principles into structured digital artefacts - including ODD definitions, evidence repositories, scenario taxonomies, and verification logic - Resembler enables contracting parties, cities, and mobility authorities to implement GRVA-level in a consistent, traceable, and repeatable manner.

Resembler Approach to Safe Deployment

The resembler system operates as a continuous governance loop consisting of four core modules: Verification Test Plan, Monitored Deployment, ODD Comparison, and Public Data Ingestion. In addition, an interactive query language provides deep analytical access to structured accident and incident intelligence.

For a user-selected geographic Operational Design Domain (ODD), the platform automatically generates a Verification Test Plan. This establishes the baseline safety case by linking deployment parameters to Nominal Scenarios and structured real-world accident and incident data derived from more than 500 million km of traffic exposure. Following approval, vehicles operate under controlled conditions, initially with a safety driver. Once stability is demonstrated, operations transition into Monitored Deployment. Operational data, deviations, and incidents are continuously evaluated against the approved baseline Verification Test Plan and declared ODD parameters.

When authorities consider expanding the ODD, modifying infrastructure, extending operating hours, or introducing software or sensor updates, the ODD Comparison Module performs a structured assessment against the previously approved configuration. Any material change automatically triggers the generation of a revised Verification Test Plan, restarting the governance cycle in a controlled and traceable manner.

Summary

The Resembler platform can be used to improve confidence level in autonomous software performance through the entire lifecycle, as shown below. During the initial stages of development, validation is done using nominal scenarios to comply with basic regulatory requirements. Prior to monitored deployment, A to B comparison of ODDs and a detailed hazard assessment in the specified ODD is necessary to improve confidence level. Finally, continuous improvement based on feedback from deployed vehicles will improve confidence level further. Using accidents as a proxy, the resembler platform provides a basis for gradual improvement in autonomous technology to ensure safe deployment.