Autonomous driving

Definition (what it is?)

Autonomous driving is the capability of a road vehicle to perform part or all of the dynamic driving task (DDT)—perception, decision-making, and control—using an onboard automated driving system (ADS) rather than continuous human operation. At higher levels of automation, the ADS, not the human, monitors the driving environment and executes object and event detection and response (OEDR) and DDT fallback (e.g., safe stop). The SAE J3016 taxonomy describes levels from Level 0 (no automation) to Level 5 (full automation in all conditions); the term “autonomous driving” is typically associated with Levels 3–5 and is distinct from driver assistance (Levels 1–2), where the human remains responsible for monitoring.

Key technical characteristics and functions

  • Sensing and perception
    • Multi-modal sensor suites: cameras (visible/NIR), radar (short/long range), lidar (mechanical/MEMS/solid-state), ultrasonic, GNSS, and inertial measurement units (IMU).
    • Sensor fusion and environment modeling: time-synchronized fusion (e.g., Bayesian/Kalman/particle filters) to build local scene models (lanes, static/dynamic objects, free space, road geometry, traffic signs/signals).
    • Robustness features: redundancy and diversity across sensing modalities; cleaning/heating for optics; algorithms for adverse weather, low light, and occlusions.
  • Localization and mapping
    • High-precision localization via GNSS with RTK/PPP, inertial dead-reckoning, wheel odometry, and vision/radar-based SLAM.
    • Use of high-definition (HD) maps where applicable; map maintenance and change detection; mapless approaches for broader generalization.
  • Prediction, planning, and decision-making
    • Prediction of other agents’ intent and motion (physics- and learning-based).
    • Behavior planning (merging, yielding, lane changes), trajectory generation, and motion planning under vehicle kinematics, comfort, and safety constraints (e.g., A, RRT, optimization/MPC).
    • Rule compliance with traffic laws and right-of-way, within the declared operational design domain (ODD).
  • Vehicle control
    • Closed-loop longitudinal and lateral control (e.g., PID, LQR, MPC) to track trajectories while managing stability, comfort, and energy use.
    • Interfaces to by-wire actuation (steering, braking, propulsion) and redundancy management for fail-operational behavior at higher levels.
  • System architecture and computing
    • High-performance compute (CPU/GPU/NPU/DSP accelerators), automotive-grade SoCs, real-time operating systems, and hypervisors.
    • Deterministic, time-synchronized in-vehicle networking (Automotive Ethernet, CAN FD, FlexRay) and precision time protocols.
    • Redundant power, compute paths, and health monitoring to tolerate faults.
  • Safety and cybersecurity
    • Functional safety engineering (ISO 26262), Safety of the Intended Functionality (SOTIF; ISO 21448), and autonomous product safety (e.g., UL 4600).
    • Hazard analysis and safety cases; minimal risk maneuvers when leaving ODD or upon failures.
    • Cybersecurity engineering (ISO/SAE 21434): secure boot, key management, intrusion detection, partitioning, and safe OTA update mechanisms.
  • Data, AI, and validation
    • Machine learning across perception, prediction, and planning; dataset curation, labeling, and bias management.
    • Continuous integration and deployment with simulation-in-the-loop, software-in-the-loop (SIL), hardware-in-the-loop (HIL), and closed-loop replay.
    • Scenario-based testing and coverage metrics for rare events; fleet data collection and synthetic data generation.
  • Connectivity and remote support (optional but common)
    • V2X (e.g., C-V2X) for cooperative perception and signaling; cloud connectivity for maps, diagnostics, analytics, and fleet learning.
    • Remote assistance/teleoperation where permitted, with clear safety envelopes and degraded-mode strategies independent of continuous connectivity.
  • Human-machine interaction
    • Clear driver role definition by level; driver monitoring systems (DMS) for Levels 1–2+; takeover requests and timing strategies for Level 3.
    • HMI design to minimize mode confusion and ensure situational awareness.

Relevance (including modern EV design)

  • Safety and efficiency: Potential to reduce collisions, smooth traffic flow, and optimize routing and driving profiles for lower energy consumption.
  • Electrification synergy: EV platforms provide high-voltage power, centralized/zonal E/E architectures, and software-defined vehicle capabilities that align well with ADS compute, by-wire actuation, and OTA updates.
  • Energy and thermal management: ADS sensors and compute impose auxiliary loads and thermal demands; EVs must budget battery capacity and cooling (e.g., liquid-cooled ECUs) without compromising range or battery life.
  • New mobility and operations: Enables robotaxis, automated delivery, depot operations, automated valet parking, and optimized charging logistics for fleets.
  • Product differentiation: Advanced ADAS and automated functions are key differentiators for both EVs and ICE vehicles, influencing user comfort, safety, and brand positioning.

Synonyms and related terms

  • Autonomous vehicle (AV), self-driving car, driverless car, robo-car, robotaxi: often used interchangeably in general discourse.
  • Automated driving / Automated driving system (ADS): the standards-preferred terminology for systems that can perform the entire DDT on a sustained basis within an ODD.
  • Advanced Driver Assistance Systems (ADAS): automate specific functions (e.g., ACC, LKA) but require continuous driver supervision and responsibility (SAE Levels 1–2).
  • Marketing terms: “Autopilot,” “Full Self-Driving,” etc., are not standards-defined and may describe Level 2 driver assistance rather than true automated driving.

Further information (regulation, ODD, safety assurance, human factors)

  • Operational Design Domain (ODD): Defines the specific conditions under which the ADS is designed to operate safely (e.g., road class, geography, weather, lighting, speed range). Performance must be validated within the declared ODD; the system must execute a minimal risk condition when exiting it.
  • Regulation and homologation: Deployment is subject to regional rules and approvals (e.g., UNECE R157 for Automated Lane Keeping Systems, national exemptions/approvals, operational permits). Compliance can include data recording, cyber requirements, and remote operation rules.
  • Validation and verification: Multi-pronged approach combining simulation at scale, proving-ground testing, on-road exposure, and scenario libraries to cover the long-tail of rare events. Safety metrics may include conflict rates, exposure-based risk, and disengagement analyses with careful interpretation.
  • Assurance and accountability: Structured safety cases, independent assessments, incident reporting, and continuous monitoring are used to manage residual risk over the product lifecycle.
  • Human factors: Clear handover protocols for conditional automation, robust DMS for driver-attentive functions, and training/user education to prevent misuse and mode confusion.

Typical hardware, materials, and manufacturing

  • Sensors and housings
    • Cameras: CMOS image sensors; glass or polymer lenses; sealed housings from aluminum or engineering plastics (PC, PA, PBT) with hydrophobic/oleophobic and anti-fog coatings; heaters or air knives for de-icing/cleaning.
    • Radar: RF-optimized PCBs and antenna-in-package; radomes from RF-transparent polymers (PC/ABS, specialized laminates); overmolded seals; corrosion-resistant finishes.
    • Lidar: Laser diodes/VCSELs, APD/SPAD receivers; coated optical glass or polymers; rotating, MEMS, or solid-state beam steering; IP6K9K-sealed enclosures, often die-cast or machined aluminum for stability and thermal control.
    • Ultrasonic: Piezoelectric transducers (PZT) in polymer overmolds integrated into bumpers/fascia.
    • GNSS/IMU: MEMS sensors in ceramic or metal packages; low-loss RF front ends; vibration-isolated mounts.
  • Compute and electronics
    • Automotive-grade SoCs with CPU/GPU/NPU; multilayer PCBs (high-Tg FR-4 or low-loss materials); thermal interface materials, heat spreaders, and liquid-cooled cold plates in aluminum/magnesium die-cast housings; EMI shielding and robust connectors.
  • Actuation and by-wire systems
    • Electric power steering with dual-redundant torque paths; brake-by-wire actuators; redundant position sensors (magnetoresistive/inductive); high-reliability relays and isolation devices for HV systems.
  • Wiring and networking
    • Shielded twisted-pair and coax for high-speed links; Automotive Ethernet backbones; sealed, EMC-rated connectors; selective use of fiber optics for very high bandwidth or EMI-sensitive paths.
  • Structural integration
    • Body-in-white and fascia optimized for sensor fields of view and RF/optical transparency; alignment features and energy-absorbing mounts; protective windows with anti-scratch coatings.
  • Manufacturing and test
    • Injection molding for radomes and housings; die casting/CNC machining for enclosures and precision brackets; cleanroom optical assembly; automated calibration (boresight and intrinsic/extrinsic) at end-of-line; environmental, shock, vibration, and sealing qualification (e.g., ISO 16750); secure provisioning of cryptographic keys for OTA and cybersecurity.

Notes and current challenges

  • Operation remains constrained by ODD limits, sensing in adverse weather, rare and ambiguous scenarios, cost and packaging of sensor/compute stacks, long-tail safety validation, regulatory variation, and public acceptance.
  • Levels describe feature capability, not the entire vehicle; a given vehicle may host multiple features at different levels. Continuous user education and clear HMI are essential to avoid misuse.