Accelerating Level-3 ADAS Validation with Virtualized SiL

Accelerating Level‑3 ADAS verification with next-generation Software‑in‑the‑Loop test systems

5 min 所要時間

The Overview

A global automotive OEM needed to validate a Level‑3 ADAS function at software speed– without relying on scarce test vehicles, limited lab time or bespoke benches. To overcome these constraints, the team re‑imagined validation as a virtual‑first, Software‑in‑the‑Loop (SiL) system, seamlessly integrated end‑to‑end with TestSphere and its Keyword‑Driven Testing (KDT) service.

The outcome was a modular, scalable validation platform that operates across laptops, datacenters or cloud platforms while integrating smoothly with existing processes and tools. This approach reduced test design effort by up to 75% and enabled sub‑day release validation for incremental builds, accelerating innovation and ensuring robust ADAS performance.

The Challenges

Traditional validation methods could not keep pace with modern development demands. Hardware‑in‑the‑Loop (HiL) and on‑road testing created bottlenecks in availability, scheduling and cost – leading to weeks‑long validation cycles even for minor software changes. To accelerate delivery, the team needed a virtualized path to shift left and scale out.

Monolithic rigs, fragmented tools: Existing approaches often “copied HIL” into SiL, importing historical constraints (real‑time only, tight coupling) and requiring testers to rewrite similar cases across tools (ECU‑TEST, TPT, Robot Framework, etc.).
Hardly any end‑to‑end traceability: Requirements, models, tests and results were scattered across different systems with manual handoffs, restricting impact analysis and reducing confidence in approval for safety‑critical ADAS features.
Evolving architecture: As vehicle platforms and toolchains advanced (with new bus technologies and simulators), the validation stack required the flexibility to swap components without requiring a complete rewrite.

The Solution

The team designed a four-component framework connected via a deterministic middleware layer, enabling plugandplay tooling and flexible “mix and match” deployments:

Simulation core
Virtual environment integrated with ego vehicle and sensors rendered in highfidelity tools (e.g., CarMaker, VTD, CARLA, SUMO) with the Open Simulation Interface (OSI) for structured sensor/vehicle/environment data. It includes a generic control API, state machine and health monitoring.
Interface provider
Decouples the simulation world from invehicle networks, hence, mapping OSI data to CAN/FlexRay/Ethernet signals and providing service mocks (e.g., backend services, domain controllers) to avoid tight coupling.
Virtual ECU (VECU)
Multiple vECU levels support different test intents – from App‑vECU for functional logic, through System‑vECU for scheduling/service communication.
Test framework
Single contact point for testers to observe/drive the system, read bus and middleware signals, orchestrate scenarios and assert outcomes –independently of the underlying simulators or ECUs.

Deterministic middleware choices (e.g., ADS2, ROS 2, DDS) are swappable per use case – supporting account tight time steps, avionicsgrade supervision or opensource ecosystems.

IT‑agnostic deployment achieved through packages built with Packer delivered as Docker images or VMs (Hyper‑V, VirtualBox) that run seamlessly on laptops, on‑prem servers or cloud (Azure/AWS/GCP).

To eliminate duplicate authoring and tool lockin, the SiL stack integrates with TestSphere, our end-to-end validation ecosystem and its KDT Service aligned to ISO 29119-5:

Write once, run anywhere: Keywordbased test designs authored once; code generator produces implementations for target tools (ECUTEST, EXAM, TPT, Robot Framework, etc.).
Composite and simple keywords: Abstract complex setup (e.g., initialize system under test, set object, ego speed) into composable; Decomposer expands these into precise steps with the correct order and signal mappings.
Sequencer and data variation: Humanreadable units (km/h, mph) are converted automatically and the parameter sweeps generate multiple cases from one design while keeping traceability back to the source requirement.
Traceability by design: Requirements, system models, test artifacts, execution logs and reports linked for stronger impact analysis and higher approval confidence.
Global keyword store: Centralized KDT database drives reuse across programs and generations, enabling oneclick fixes to cascade across thousands of cases.

The Impact

Up to 75% reduction in test design time compared with conventional, tool‑specific authoring – based on multiyear KDT adoption metrics.
Release validation in approximately one day for incremental builds, replacing weeklong cycles and accelerating OTAready software delivery.
Faster toolchain evolution: Porting the simulation core between vendors/opensource stacks completed by two engineers in under a month, preserving prior work and minimizing retraining.
Scale on demand: Run anywhere – from developer laptops to cloud – via container/VM images and CI orchestration; spawn parallel nodes to expand scenario coverage rapidly.
Higher quality and compliance readiness: Builtin determinism, health monitoring and endtoend traceability improved reviewability for safety assessments and made impactbased regression practical.

Highlights

Scope: Level‑3 ADAS validation via software‑in‑the‑loop, integrated with TestSphere plus KDT.
Architecture: Simulation core ↔ Interface provider ↔ VECU ↔ Test framework; deterministic middleware (ROS 2 / ADS‑2 / DDS).
Deployment: Docker/VM images via Packer; laptop, on‑prem or cloud (AWS /Azure/GCP).
Tooling: CarMaker / VTD / CARLA / SUMO; ECU‑TEST / EXAM / TPT / Robot Framework; FMI/FMU and SILVER integrations.
Results: Up to 75% design effort reduction; less in 1‑day release validation; rapid simulator swaps; scalable scenario execution.

Why this approach works

Virtualization is now a critical enabler for Software‑defined Vehicles (SDVs). By decoupling simulation, interfaces and vECUs – and by abstracting test design with KDT – the organization gains speed, scale and repeatability without discarding existing tools. The SiL platform creates a reusable validation asset that evolves with architecture changes and supports a continuous flow of ADAS software releases. The SiL platform creates a reusable validation asset that evolves with architecture changes and supports a continuous flow of ADAS software releases.