kineton ai guardian cover title IT
AI-GUARDIAN

Continuous, non-invasive monitoring of Advanced Driver Assistance Systems
in real-world conditions.

AI-Guardian 3D representation

Where traditional diagnostics detect failures, AI-GUARDIAN detects behavioral anomalies, real-world deviations that no fault code will ever surface.

The Challenge

When everything works, but something still goes wrong.

Modern ADAS systems pass rigorous validation protocols, yet real-world incidents continue to occur. The gap lies not in hardware failures, but in behavioral anomalies: edge cases where the system operates within diagnostic tolerances but produces unexpected, potentially unsafe responses.

Traditional on-board diagnostics (OBD) detect component failures and electrical faults. They monitor signals against fixed thresholds and report Diagnostic Trouble Codes when a clear error occurs. But what happens when all components are technically functional, yet the system's overall behavior deviates from what is safe and expected in a specific real-world context?

This is precisely the challenge addressed by SOTIF, Safety Of The Intended Functionality (ISO 21448). SOTIF recognizes that hazardous situations can arise not from system failures, but from performance limitations and foreseeable misuse in complex, unpredictable environments. AI-GUARDIAN was built to close this monitoring gap.

SOTIF / ISO 21448
Diagnostic gap

An Adaptive Cruise Control that doesn't trigger any fault but keeps losing the track of a leading vehicle in a curvy road making a series of undesired acceleration and decelerations. This unexpected behaviour is invisible to traditional diagnostics. AI-GUARDIAN sees it

The behavioral layer

Where OBD monitors components, AI-GUARDIAN monitors intent vs. outcome , analyzing whether the system behaves as it should across the full diversity of real-world scenarios.

The Solution

Monitoring ADAS beyond traditional diagnostics

AI-GUARDIAN integrates Edge AI, real-time vehicle data, and Cloud Computing into a unified behavioral monitoring framework, purpose-built for the complexity of modern ADAS.

Deployed at the edge, directly within the vehicle, AI-GUARDIAN continuously reads signals from the vehicle's communication networks (CAN, FlexRay, Ethernet) without requiring any modification to existing ECUs. An onboard AI engine, trained on both field data and physical system models, establishes the expected behavioral envelope for each ADAS function under analysis.

When deviations are detected, relevant data is securely transmitted to a cloud analytics layer, where deeper root-cause analysis correlates behavioral symptoms with environmental context, sensor inputs, and historical patterns. The result: actionable intelligence, not just alerts.

Edge AI
Vehicle Network Data
Cloud Analytics
Physical Models
Real-Time Monitoring
How It Works

Four integrated phases of intelligent monitoring

01 — MONITOR

Monitor

On-board Edge AI analyzes vehicle network signals and detects deviations from expected ADAS behavior, in real time, without driver intervention.

02 — DETECT

Detect

Behavioral anomalies are identified in real-world driving conditions, the kind of edge cases that validation environments cannot fully anticipate or reproduce.

03 — EXPLAIN

Explain

Cloud-based analytics correlate symptoms with potential root causes, linking behavioral deviations to environmental conditions, sensor limitations, or interaction patterns.

04 — LEARN

Learn

Real-world edge cases become enriched datasets for continuous AI model improvement — creating a feedback loop that makes ADAS systems progressively safer over their operational lifetime.

Why AI-GUARDIAN

Designed for what traditional tools can't see

◻

Non-invasive

Works with standard vehicle network signals without modifying existing ECUs or requiring hardware changes to the vehicle architecture. Integration is fast, risk-free, and transparent.

◻

AI-powered

Uses a combination of machine learning algorithms and physical system models to identify unexpected ADAS behavior, going beyond pattern matching to true behavioral understanding.

◻

Root-cause analysis

Helps distinguish between environmental conditions, human behavior, and system limitations as contributors to anomalous events, providing context, not just alerts.

◻

Continuous learning

Data collected from real-world anomalies is transformed into structured datasets for OEMs AI models and system development improvement, enabling ADAS safety to evolve with the full diversity of real driving scenarios.

From Validation to Continuous Monitoring
"ADAS safety doesn't end when validation does."

Lifecycle safety for the real world

Validation is a milestone, not the finish line. Once a vehicle leaves the controlled environment of a test track or simulation framework, it enters a world of infinite variability: changing weather, unexpected obstacles, novel road geometries, and unpredictable human behavior.

AI-GUARDIAN reframes ADAS safety as a continuous operational discipline. By monitoring system behavior across the entire vehicle lifecycle, from initial deployment through years of field operation, it ensures that safety performance can be tracked, measured, and improved over time.

This approach aligns directly with the emerging regulatory landscape around functional safety (ISO 26262) and SOTIF (ISO 21448), which increasingly recognize that post-market monitoring and field data collection are essential components of a complete safety strategy, not optional add-ons.

Technology

Core technology stack

AI-GUARDIAN is built on a multi-layer technology architecture combining embedded intelligence, vehicle communication protocols, and scalable cloud analytics.

Edge Computing Artificial Intelligence Physical Models CAN / FlexRay / Ethernet Cloud Computing MLOps Root-Cause Analysis Continuous Learning
Application & Value

Built for the next generation of mobility

AI-GUARDIAN delivers measurable value across the automotive value chain, from engineering teams working on ADAS development to end users who depend on these systems every day.

For OEMs & Tier-1 Suppliers

  • Accelerate ADAS validation with real-world behavioral data
  • Identify SOTIF-relevant edge cases before they become incidents
  • Support ISO 21448 and ISO 26262 compliance documentation
  • Reduce warranty costs through early anomaly detection
  • Build a continuous feedback loop for AI model improvement
  • Enable over-the-air (OTA) safety updates grounded in field evidence

For Safer Mobility

  • Ensure ADAS systems perform safely across all real-world conditions
  • Bridge the gap between lab validation and field performance
  • Provide transparency into how and why ADAS decisions are made
  • Support regulatory requirements for post-market surveillance
  • Contribute to shared industry knowledge on ADAS edge cases
  • Protect end users through continuous, invisible monitoring

What happens when an ADAS behaves unexpectedly?

AI-GUARDIAN helps you understand why.

“AI Guardian - AI-Powered Platform for Behavioral Anomaly Detection
and Reliability Monitoring of Advanced Driver Assistance Systems”

CUP: B69J25000600005

Co-financed by the European Union – Italian Republic – Coesione Italia 21-27, Research, Innovation and Competitiveness

Grant Reference

FONDO CRESCITA SOSTENIBILE – PN RIC 2021/2027 “STEP” – Action 1.1.4.
“Collaborative Research”, pursuant to Ministerial Decree of 25 October 2024 – Director’s Decree of 3 April 2025

Scroll to Top