Autonomous Driving Responsibility Boundary

Definition

Autonomous driving responsibility boundary is the distinction between systems that assist a human driver and systems that assume the driving task. It asks who must monitor, intervene, own fallback, carry liability, and explain failures when driving automation is active.

Current Synthesis

The bounded sources agree that responsibility is more important than feature names or scene coverage. L2 assists but leaves the driver responsible, L3 creates a conditional handoff state, and L4 moves fallback responsibility to the system and operator so riders become passengers. Current mass-market systems can feel capable and useful without crossing that boundary.

Regulation is part of the product definition, not only post-accident adjudication. Clearer L3 obligations can bind a manufacturer’s claim to responsibility, reduce ambiguous “L2.x” marketing, and help users calibrate supervision. Trust and natural behavior can increase adoption, but neither silently transfers legal fallback.

Key Claims

• Autonomy levels should be read through responsibility and fallback, not only through feature richness.
• L2 remains a driver-responsibility product even when it includes lane keeping, following, LCC, NOA-style navigation, or other advanced assistance.
• L3 is commercially and legally hard because control can move back to a human exactly when attention has decayed.
• L4 creates a different service and operating model because the passenger is no longer the fallback driver.
• Mass-market assisted-driving scale and user trust do not themselves change responsibility.
• “L2.x” capability labels do not create a new responsibility level between L2 and L3.
• Regulation, safety evidence, handoff rules, driver education, and operating design have to change together before responsibility shifts.

Evidence

• L2/L3/L4 taxonomy: 没有方向盘的出行，走到哪一步了？ NVIDIA × 小马智行一次聊透智能驾驶 uses 张宁’s account to distinguish driver-responsible L2, handoff-ambiguous L3, and system/operator-responsible L4.
• Robotaxi operations: 没有方向盘的出行，走到哪一步了？ NVIDIA × 小马智行一次聊透智能驾驶 links L4 responsibility to charging, cleaning, maintenance, dispatch, accident response, towing, passenger support, and fleet lifecycle operations.
• Mass-market boundary: No.215 华为不造车，鸿蒙智行到底是什么？ says current Chinese-market assisted driving still leaves the legally responsible driver as human while standards may create a clearer future path.
• Trust versus responsibility: EP279 当方向盘慢慢松开，我们如何与车相处？ shows owners becoming more comfortable with assisted driving through lived experience, while Peng Lei still frames future higher-order convenience as dependent on clearer regulation.
• Legal and marketing boundary: 特番｜从会开车到会做事：AI如何进入真实世界 says current mainstream consumer use remains L2 despite broader city and highway coverage, and argues that clearer L3 law can attach responsibility to the manufacturer’s claim while limiting ambiguous “L2.x” language.

Counterevidence & Qualifications

• The sources do not provide a full legal analysis of Chinese, U.S., or European liability regimes.
• Owner comfort is not equivalent to validated system responsibility.
• L3 and L4 terminology can be used differently in marketing, regulation, and engineering contexts, so the page treats fallback as the stabilizing distinction.
• Assisted-driving adoption can be socially valuable even when responsibility remains with the human driver.
• The newest source discusses pending or proposed Chinese road-law clarification but is not a primary legal text; exact statutory effects remain source-scoped.

What Changed

• Added regulation as an ex-ante product-claim discipline, not only an accident-liability mechanism.
• Clarified that broader scene coverage and “L2.x” branding do not themselves change fallback responsibility.

Related Concepts

• Assisted Driving Trust Formation / 辅助驾驶信任形成 - user adoption process that must not blur fallback responsibility.
• Robotaxi Economics - business consequences of removing the driver.
• Robotaxi Fleet Operations - operational responsibilities that replace driver labor.
• Autonomous Vehicle Safety Benchmark - safety-evidence frame needed when the system becomes responsible.
• Autonomous Vehicle Regulatory Patchwork - legal and policy context around L4 deployment.
• Autonomous Driving Data Flywheel - deployment-scale branch that can improve systems but does not itself shift liability.