ADAS and automated driving systems: Engineering Safety, Failure Analysis and Validation

Introduction and application problem 

An automated emergency braking function may execute its software correctly and still miss a partly occluded pedestrian or misinterpret a roadworks scene. A camera power fault or corrupted control message is a different, malfunctioning-behaviour problem. ADAS engineering must separate these classes while demonstrating how the vehicle detects limitations, chooses a fallback and informs the driver. ISO 26262, SOTIF, scenario validation and cybersecurity each contribute different evidence; none alone proves safety in every road condition.

  • ADAS can directly influence braking, steering and acceleration; false or missed intervention can create hazards. 
  • Perception depends on weather, lighting, occlusion, road geometry, calibration and object appearance, not only electronic fault status. 
  • Driver supervision and handover assumptions change with automation design and must be explicit in the ODD. 
  • Sensor timing, compute scheduling, maps, networks and actuators create cross-system dependencies. 
  • Updates and remote data can change behaviour after release, requiring configuration and cybersecurity control. 

  • Define feature, ODD, driver role, fallback concept and vehicle boundaries before requirements. 
  • Use HARA/ISO 26262 for malfunctioning behaviour; use SOTIF analysis for functional insufficiencies and triggering conditions; perform TARA where relevant. 
  • Specify safety goals and degraded/fallback behaviour for sensor, compute, actuator faults and ODD exit. 
  • Build a risk-informed scenario catalogue using field data, simulation, expert review and known edge cases. 
  • Combine software/hardware verification, simulation, proving-ground and controlled vehicle tests with explicit release criteria. 

Application-specific safety and security challenges 

  • Adverse weather and visibility: Rain, fog, glare, darkness or dirty sensor windows can degrade detection without a clear hardware fault; validate limitation detection and transitions. 
  • Occlusion and unusual road users: Partially hidden pedestrians, cyclists and atypical objects challenge perception; test missed detection and unsafe false positives. 
  • ODD transitions and handover: Roadworks, lane loss or map mismatch can leave the validated domain; verify detection lead time and fallback. 
  • Rare combinations and fleet variation: Road geometry, loading, friction and sensor/software variants complicate coverage; document scenario rationale and simulation correlation. 

Application architecture and operating context 

The boundary can include sensors and cleaning systems, time synchronisation, fusion, perception, prediction/planning, vehicle-motion inputs, central/zonal compute, networks, brake/steering/propulsion interfaces, HMI, driver monitoring, maps and remote inputs. Specify ODD, fallback ownership, sensor-health indications, timing assumptions and driver role. For higher automation, distinguish a warning from a system-controlled fallback; do not assume a human can always recover immediately. 

Applicable standards and boundaries 

  • ISO 21448:2022 (SOTIF): functional insufficiencies and triggering conditions for situational-awareness-dependent functions; ISO catalogue indicates revision activity, so recheck status. 
  • SAE J3016: driving-automation terminology/taxonomy; not a functional-safety certification standard. 
  • ISO 26262: malfunctioning E/E behaviour. 
  • ISO/SAE 21434 and UN R155: cybersecurity engineering and regulatory context where applicable. 
  • UN R156 and ISO 24089: software update management/engineering where applicable. 
  • Function-specific vehicle regulations for braking, steering and other ADAS features must be selected by market and vehicle category. 

Applicability note: this is a screening set, not a declaration that every standard applies to every product. Confirm the exact item, vehicle category, market, contractual requirements and current edition/amendment status. A draft standard must not be presented as a published requirement. 

Application safety lifecycle 

  1. Item definition and context: define intended use, interfaces, modes, assumptions, users and external dependencies. 
  1. Hazard analysis and safety goals: identify hazardous vehicle-level behaviour and derive goals through the applicable process; do not invent an ASIL. 
  1. Functional and technical safety concepts: specify required behaviour, fault reactions, safe/degraded states and interface assumptions. 
  1. Architecture and implementation: allocate requirements to hardware, software, mechanical/electrical elements and operations; analyse independence and common causes. 
  1. Safety analysis and integration: maintain traceable FMEA/FMEDA, fault trees and dependent-failure analysis where appropriate, with assumptions and data sources. 
  1. Verification and validation: combine reviews, analysis, unit/integration tests, HIL, fault injection, environmental testing and vehicle/application validation as relevant. 
  1. Production, service and change: control versions, calibration, supplier changes, repair, software updates, field feedback and change-impact analysis. 

Hazards, failure scenarios and safety objectives 

The scenarios below are illustrative engineering examples, not a project HARA, assigned ASIL, approved safety goal or quantitative risk assessment. Final classification depends on the defined vehicle item, operating situations and applicable safety process. 

Illustrative hazardous behaviour  Possible initiating fault / condition  Potential consequence  Example analysis or evidence 
Unintended emergency braking  False classification, software fault or corrupted target data  Rear-end collision or loss of control  Fault analysis, false-positive scenarios and actuator monitoring 
Pedestrian not detected  Occlusion, low contrast or perception limitation  Collision with vulnerable road user  SOTIF scenarios and end-to-end validation 
Lane keeping steers toward edge  Lane-model/calibration fault or stale vehicle state  Road departure  Fault injection and roadwork/lane-marking tests 
Feature active outside ODD  Boundary detection/HMI fault  Driver relies on unavailable assistance  ODD-exit and fallback timing tests 

 

Safety requirements, architecture and implementation 

Specify function, ODD, input quality, timing, diagnostics, actuation limits, fallback trigger, driver indication and recovery. For SOTIF, define scenario classes, acceptance criteria and residual-risk argument. Do not claim simulation mileage proves safety unless scenario distribution, model realism, oracle and correlation assumptions are justified. ASIL and performance targets must follow the specific safety process. 

Verification, validation and safety evidence 

Use sensor/interface timing tests, fault injection, network load and compute stress, HIL, actuator response, simulation sweeps, sensor-in-the-loop, proving-ground and vehicle testing. Track false positives/negatives by scenario, ODD boundary detection and fallback performance. Preserve software/model/data versions, calibration, scenario definitions and test oracles. Evidence is bounded by the scenarios and configurations exercised. 

Common pitfalls and how to detect them 

Pitfall  Why it matters  Detection method / evidence  Better engineering practice 
Treating SOTIF and ISO 26262 as interchangeable  Fault-free functionality can still be insufficient  Review hazard taxonomy and safety case  Maintain linked but distinct analyses 
Defining ODD only as prose  Ambiguous limits cannot drive reliable exit detection  Check ODD terms against measurable conditions  Create testable ODD entry/exit requirements 
Counting simulation volume instead of coverage  Large runs may omit edge cases or use unrealistic models  Audit scenario diversity, oracle and correlation  Use risk-informed complementary evidence 
Assuming takeover always succeeds  Driver availability and reaction are not guaranteed  Human-factors and fallback tests  Define fallback appropriate to automation design 

 

Relationship to adjacent disciplines 

Functional safety addresses hazards caused by malfunctioning behaviour of E/E systems. Reliability engineering helps characterise failures and lifetime; cybersecurity addresses malicious threats; SOTIF is relevant where intended-functionality insufficiencies create risk; quality/process assurance supports consistent development; human factors addresses user interpretation; and environmental/electrical/homologation activities address other requirements. These disciplines exchange assumptions and evidence but are not interchangeable. Select only those relevant to the application and document their interfaces. 

  • ☐ Technical reviewer checked engineering argument, hazard pathways, lifecycle logic and illustrative examples. 
  • ☐ Standards reviewer verified current edition/status, scope, terminology and regulatory applicability. 
  • ☐ VerveTronics verified service, qualification, tool, experience and delivery claims. 
  • ☐ Marketing/editorial reviewer checked keyword ownership, search intent, metadata, canonical URL and overlap. 
  • ☐ CTA and internal links are live; approved visuals, captions, alt text and structured data are correct. 
  • ☐ No invented customer case studies, measured results, certification outcomes or unsupported numeric claims. 
  • ☐ Any real case study/customer detail has appropriate permission and anonymisation. 

  • Camera/radar/lidar or vehicle-state sensor faults, misalignment, contamination and time mismatch. 
  • Functional insufficiencies: unusual objects, cut-ins, roadworks, glare, low contrast, adverse weather and poor confidence estimates. 
  • Unintended braking/steering, delayed torque request or mismatch between commanded and achieved deceleration. 
  • Feature remains active outside the ODD; unclear status, late takeover request or mode confusion. 
  • Compute overload, thermal throttling, memory corruption, network loss and common-cause faults. 
  • Threats to sensor data, messages, maps, updates or remote control paths. 

VerveTronics describes automotive functional-safety engineering across ADAS, safety concepts, HARA, system/hardware/software analyses and verification/validation. For an ADAS programme, scope should distinguish fault-based safety from SOTIF scenarios, ODD assumptions, fallback design and cybersecurity interfaces. No single standard or test campaign proves safety in every environment. 

VerveTronics automotive functional safety and ISO 26262 services

  • Is ADAS safety only an ISO 26262 problem? No. ISO 26262 addresses malfunctioning behaviour; ISO 21448 is relevant to functional insufficiencies, while cybersecurity and vehicle regulations may also apply. 
  • Does ASIL prove safe perception in all weather? No. ASIL is part of fault-based safety engineering and does not replace ODD/SOTIF validation. 
  • Can simulation replace vehicle tests? Simulation scales scenario exploration, but models, oracles and correlation need justification and complementary validation.