Autonomous driving has been the hottest technology race in the automotive industry since 2016. But “autonomous driving” is frequently overhyped — actual product technical maturity lags significantly behind public perception. Understanding SAE’s (Society of Automotive Engineers) L0-L5 classification framework is the baseline for evaluating autonomous driving progress.
## SAE Level Breakdown
**L0 (No automation)**: driver has complete control; system may provide warnings but cannot intervene in control. Examples: lane departure warning, forward collision warning.
**L1 (Driver assistance)**: system controls either steering or speed, but not both; driver handles the other. Examples: adaptive cruise control (ACC) or lane keeping assist (LKA) — but only one at a time.
**L2 (Partial automation)**: system simultaneously controls steering and speed, but driver must continuously monitor and be ready to take over. Tesla Autopilot, Xpeng XPILOT, and AITO HUAWEI ADS are all L2. Key concept: driver-in-loop — hands can leave the wheel but eyes cannot leave the road.
**L2+ (Enhanced L2)**: an informal industry classification for advanced driver assistance with auto lane change, highway navigation assist, etc. — Tesla FSD (China version), Xpeng XNGP, Huawei ADS 3.0. Technically still L2 but experientially close to L3.
**L3 (Conditional automation)**: under specific conditions (highway, low-speed traffic jam) the system drives entirely; the driver may disengage but must quickly take over when the system requests. Mercedes-Benz Drive Pilot is the world’s first regulation-certified L3 system (Germany highway, ≤60km/h). China’s regulations opened L3 road access in 2023 (pilot in Shenzhen, Shanghai). L3’s core barrier is **liability attribution**: in an accident, who is responsible — the driver or the automaker?
**L4 (High automation)**: within a specific geographic area, no driver needed; the system handles all situations. Baidu Apollo, WeRide, DiDi Autonomous Driving, and Waymo (US) robotaxis are L4. Baidu’s Apollo Go operates fully driverless (no safety driver) commercial service in Wuhan, with peak daily orders exceeding 34,000 in 2024.
**L5 (Full automation)**: any road condition, any scenario, no human driver needed. No mass-production product has reached this standard; academic consensus on an L5 path has not yet formed.
## The Technical Debate: LiDAR vs. Pure Vision
Tesla’s pure vision approach (cameras only) versus Baidu, Waymo, and Xpeng’s multi-sensor fusion (LiDAR + millimeter-wave radar + cameras) represents the field’s largest ongoing technical dispute. Tesla argues LiDAR is “expensive and redundant” — humans drive with eyes (cameras) alone; critics argue cameras have insufficient reliability in extreme weather and at night.
See [In-Vehicle LLMs and Intelligent Cockpits](https://sunqi.org/car-large-language-model-en/) and [SAE autonomous driving level standard](https://www.sae.org/standards/content/j3016_202104/).
