AI-Powered E2E Automotive Testing Framework for In-Vehicle Voice and Infotainment Systems

E2E Testing Framework is an Intellias’ AI-enabled solution that automates end-to-end automotive infotainment testing of in-vehicle voice assistance systems and digital cockpits, from requirements to nightly regression on head units in the lab.

With every release, digital cockpit teams have more voice, navigation, and media functions to retest. The most important scenarios span several systems: a driver’s request has to produce the right answer, screen, and action in the vehicle. Component tests can pass while that journey fails, and scripted automation breaks when voice replies vary in wording. As a result, these checks often stay manual, and regression slows down releases. Where infotainment testing automation is in place, device faults show up as failed tests, and results rarely carry enough evidence to tell a product defect from lab noise, so engineers recheck runs by hand.

The automotive infotainment testing framework automates these journeys end to end. AI agents turn requirements and manual test cases into tests, or testers describe scenarios in plain language. Every night, the framework speaks to the head unit, grades the answer, screen, action, and voice quality separately, and stores the evidence behind each verdict. Reports are ready before engineers start their day.

AI-enabled infotainment testing from requirement to head unit

Tests can start in two ways. AI agents read user stories, acceptance criteria, and existing manual test cases from Jira, derive test cases, and write the runnable scenario. Each generated test passes a hardware-free review gate, and an engineer approves it before merge. The in-vehicle infotainment testing framework can also explore the head-unit interface on its own and discover key controls and screens. Alternatively, a testing engineer can write the scenario in plain language, and the framework turns it into a voice prompt and a set of checks

During a run, text-to-speech (TTS) delivers the wake word and command to trigger the head unit’s own voice assistant, and automatic speech recognition (ASR) transcribes the reply. Each infotainment automation testing run then receives four verdicts. An LLM judge scores the answer against the intent of the request. Screen and action verdicts rely on deterministic checks of the UI hierarchy and system state through UI Automator 2, for example, that a route was created. Where no deterministic check can reach the screen, a vision-language model (VLM) reads it. Voice quality is graded as a separate verdict.

Every test run saves audio, video, screenshots, and full run data, so each verdict can be reviewed later.

Engineers stay in control of what reaches the nightly suite. A new test joins it only after it passes consistently on a debug device. Before each nightly run, a preflight step brings up the head unit through the lab relay box and resets it to a clean state. An unresponsive device is reported as an infrastructure issue and kept out of test results. Infrastructure retries are tracked separately, and the test design owner triages the remaining failures into test defects and product defects.

Solution goals

Validate the full driver journey – Check the spoken answer, the screen, and the resulting system action in one automated scenario.

Make test creation accessible – Let testers describe scenarios in plain language or generate them from existing requirements and manual test cases.

Deliver verdicts teams can verify – Keep infrastructure faults out of defect reports and store the evidence behind every result.

Run regression every night – Prepare results before the working day begins.

Main features of the framework

Modern benefits operators need modern engineering. We build it.

Four verdicts per run – Answer, screen, action, and voice quality are graded separately, so a report shows which part of the journey failed.

LegalTech Solutions

Autonomous UI exploration – The framework navigates the head-unit interface on its own and discovers key controls and screens.

AI-Powered E2E Automotive Testing Framework

Full evidence capture – Every run stores audio, video, screenshots, full run data, and an Allure report.

AI-Accelerated Delivery

Requirement traceability – Acceptance criteria flow in from Jira, and test cases and results flow out to XRAY, so each test stays linked to the requirement it covers.

Digital Healthcare Solutions

Lab Relay Box and preflight checks – Each nightly run starts with device bring-up and a clean state, and device faults are reported separately from test results.

Modernizing Planstin's Benefits Administration Platform for Secure, Scalable Growth

Separate development and regression flows – New tests go through their own pipeline, and the framework is designed to run development and nightly regression on separate device pools.

Modernizing Planstin's Benefits Administration Platform for Secure, Scalable Growth

Modular skills – New capabilities are added as skills, so the framework can extend to other vehicle domains, mobile apps, and web interfaces.

Business value for automotive OEMs and suppliers

Overnight regression
of voice journeys

Voice scenarios that previously required manual testing run automatically every night. Regression results for each build are known by morning.

Defects caught across layers
before vehicle testing

Checking the answer, the screen, and the system action together exposes integration defects that component tests miss. Teams find them in the lab, before in-vehicle validation, where fixes take longer and cost more.

Verdicts backed
by evidence

Every failure comes with the audio, video, screenshots, and run data behind it. Developers can review what happened in the run before trying to reproduce the defect on hardware.

Results engineers
can act on

Preflight checks, test stabilization, and separate tracking of infrastructure retries keep device faults out of defect reports. Engineers spend less time rechecking the lab setup.

Manual effort shifts component by component

Manual regression for a component stops only after its nightly runs stay green and the test design owner accepts the results. One engineer owns test design and acceptance for the automated suite, while the QA team moves to new and safety-related features, exploratory and UX testing, real-world driving tests, and helping developers reproduce complex defects.

Traceability from requirement to result

Each acceptance criterion is linked to its test case and its latest XRAY execution, which gives QA management a direct view of coverage and results.

Bring E2E in-vehicle infotainment system testing to your digital cockpit

Connect with Intellias to explore how AI-enabled car infotainment system testing can cover voice, display, and navigation journeys in your vehicle programs.

Connect
AI-Powered End-to-End Testing Framework 
for In-Vehicle Voice and Infotainment Systems image

Voice navigation test use case scenario

Task

A test team needs to verify that the voice assistant handles a navigation request correctly: the voice assistant answers, the screen shows the destination, and the head unit sets the correct route.

Key steps
01

Test design

An AI agent derives the test from the Jira user story and acceptance criteria, or a tester writes it in plain language. After a hardware-free review gate, an engineer approves it.

02

Stabilization

The test runs on a debug device until it passes consistently, then joins the nightly suite.

03

Preflight

The head unit is brought up through the lab relay box and reset to a clean state.

04

Voice prompt

TTS speaks the wake word and a navigation request to the head unit.

05

Answer verdict

ASR transcribes the reply, and the LLM judge scores it against the intent of the request.

06

Screen and action verdicts

UI Automator 2 confirms that the correct route is set, and a VLM checks the navigation screen where needed.

07

Publishing

The evidence bundle and Allure report are uploaded, results are published as an XRAY test execution for the build, and confirmed product defects are logged in Jira.

What matters most

Each verdict is reported separately. A correct spoken answer without a route shows up as a failed action verdict, so the report points to the part of the journey that broke.

Result

The team receives a verified result for the complete voice-to-route journey with every nightly build, together with the evidence needed to review any failure.

System architecture

The automotive infotainment system testing framework connects requirements, lab hardware, and defect tracking in one automated flow. At design time, AI agents turn requirements into tests in the framework repository, with a hardware-free review gate in Azure DevOps. At run time, a nightly Azure DevOps pipeline picks up the latest application build from GitLab, and a thin test harness drives the head unit through the lab relay box. A dedicated GPU host runs TTS, ASR, the LLM judge, and the VLM, so no models run on the CI agent. Evidence and Allure reports are stored in Azure Blob Storage, and results are published to XRAY.

HIGH-level architecture
HIGH-level architecture
Technology stack
Python TTS ASR LLM judge VLM UI Automator 2 ADB NVIDIA Jetson AGX Thor Azure DevOps GitLab Azure Blob Storage Allure Jira Confluence XRAY Lab Relay Box