Recently, the MIIT officially released China's first mandatory national standard for L3 and L4 high-level autonomous driving, titled "Safety Requirements for Autonomous Driving Systems in Intelligent Connected Vehicles." The standard will take effect on July 1, 2027, requiring all newly declared vehicle models equipped with L3/L4 autonomous driving to meet all mandatory provisions. Existing models will be granted a one-year transition period for compliance adjustments.

Upgraded from the previous voluntary national standard GB/T 44721-2024, this standard ends the era of unregulated growth in China's high-level autonomous driving sector, characterized by "no mandatory market access rules, ambiguous human-machine responsibility, and fragmented regional pilot standards." It establishes national-level safety baselines for the full lifecycle of L3 (conditional autonomy) and L4 (high autonomy) systems, serving as a core regulatory framework linking technological R&D, commercial operations, and industry oversight.

Unlike previous voluntary guidelines focusing on technical recommendations, this national standard carries legal enforceability. It covers all passenger and commercial vehicles across operational and consumer-grade autonomous driving categories, defining five core systems: operational boundaries, human-machine role allocation, failure emergency response, corporate safety management, and testing/acceptance protocols. Supporting mechanisms include safety documentation, data retention, and full-process verification. For the industry, this document transcends technical testing specifications—it represents top-level design to reshape competition, resolve commercialization barriers, and improve (wán shàn - improve/perfect) supporting legal frameworks.

Over the past five years, China's intelligent connected vehicle (ICV) industry has achieved leapfrog development. L3 vehicles have received conditional road approvals, while L4 autonomous driving has entered normalization (cháng huà - regularized) pilot operations across ports, mines, industrial parks, trunk logistics, and urban Robotaxi services. Over 100 cities nationwide have opened autonomous driving road testing, driving continuous market expansion. However, the industry has long relied solely on voluntary standards and local pilot management measures, creating three major pain points:

First, rampant grade inflation in marketing. Many automakers package L2 driver assistance as "quasi-L3/fully autonomous driving," deliberately blurring SAE classification definitions. Some models lack hardware redundancy and proper handover logic, relying on marketing rhetoric to mislead consumers.

Second, no unified basis for human-machine accident liability. Accidents frequently occur when L3 systems exceed operational boundaries or drivers fail to take over promptly, and when L4 operational vehicles experience sudden failures. Without national standards defining liability boundaries, accident dispute resolution and insurance claims have long been deadlocked.

Third, inconsistent regional pilot regulations. First-tier cities and third/fourth-tier cities impose vastly different requirements for autonomous driving road access, safety officer deployment, and testing mileage. High cross-regional operational costs hinder the formation of a unified national replication model.

Meanwhile, global autonomous driving safety incidents have surged, with multiple overseas cities tightening Robotaxi operational restrictions. This proves that high-level autonomous driving without rigid safety constraints poses significant road traffic hazards. At this critical juncture of transitioning from pilot projects to mass production, implementing mandatory unified safety standards is essential to safeguard public travel safety and regulate industry development.

Currently, Europe, the U.S., Japan, and South Korea have all introduced mandatory regulations for high-level autonomous driving, while the United Nations is advancing the Global Technical Regulation (GTR) for autonomous vehicles. National standard systems are intensifying competition. China's standard development process fully aligns with ISO/SAE international frameworks while incorporating domestic complex road conditions and urban-rural traffic environment differences, forming a safety indicator system adapted to local conditions.

After implementation, China will possess a complete, independently controllable L3/L4 market access, testing, and safety management framework. This provides a basis for standard mutual recognition for Chinese automakers going global and enables continuous export of China's autonomous driving governance experience, enhancing international industry influence.

In recent years, China has continuously issued top-level policies for intelligent connected vehicles, from the "New Energy Vehicle Industry Development Plan" to autonomous driving road test access rules, gradually constructing a four-tier governance framework of "top-level planning - local pilots - mandatory national standards - supporting regulations." This mandatory national standard fills institutional gaps in market access for mass-produced autonomous vehicles, linking upstream component vehicle-grade standards with downstream regulations on road traffic, insurance, and traffic liability determination, enabling full-chain supervision from R&D to operation.

The mandatory standard comprises four core sections: basic safety capability requirements for autonomous driving systems, human-machine interaction and takeover management rules, minimum risk mitigation strategies (MRM) for failures, and enterprise full-lifecycle safety assurance and inspection mechanisms. All provisions include quantifiable, testable hard indicators without ambiguous flexible wording.

1. Clearly define Operational Design Domain (ODD) to establish system capability boundaries

The Operational Design Domain (ODD) forms the basis for distinguishing autonomous driving capabilities and allocating safety responsibilities. The standard mandates automakers to fully disclose all applicable boundaries for L3 and L4 systems across six dimensions: road type, speed range, weather visibility, lighting conditions, traffic flow density, and road surface status. The system must continuously verify whether it operates within legal ODD boundaries in real-time.

When vehicles approach preset operational boundaries or encounter conditions exceeding system capabilities (e.g., rain/fog low visibility, icy roads), graded warnings must activate in advance, prohibiting continued autonomous operation under insufficient capability conditions. Special quantitative indicators are set for L3 highway scenarios: minimum forward perception detection distance ≥130 meters at 120km/h cruise speed, lateral perception covering ±9 meters from the vehicle body. In complex environments (light rain, night without streetlights), perception detection distance attenuation must not exceed 20%, forcing multi-sensor fusion solutions to address single-vision perception vulnerabilities in harsh conditions.

2. Tiered human-machine interaction design with rigid constraints on L3 takeover rules

The standard establishes two completely independent human-machine interaction logics for L3 and L4, with the core distinction being whether human backup drivers are required.

L3 Conditional Autonomous Driving: Driver as legal ultimate responsible party

The standard mandates multi-dimensional Driver Monitoring Systems (DMS), requiring at least two independent indicators beyond steering wheel torque (e.g., seat occupancy, seatbelt status, eye gaze, head posture). Immediate Level 1 audible/visual warnings activate when drivers leave seats, become distracted, close eyes, or unfasten seatbelts. If effective takeover status isn't restored within 10 seconds, warnings escalate to multi-level reminders including seatbelt pretension and seat vibration. The system must reserve a minimum 10-second complete takeover window to eliminate emergency takeover risks.

Liability is clearly divided: when accidents occur within ODD boundaries without takeover requests, automakers bear primary responsibility. If accidents result from delayed manual takeover after system warnings, drivers assume corresponding liability, resolving long-standing human-machine accident liability disputes.

L4 High Autonomous Driving: System independently handles all dynamic driving tasks

The core rule for L4 is no reliance on occupant takeover. Even for vehicles equipped with steering wheels, passengers have no legal takeover obligations. Steering-wheel-free Robobuses, mining autonomous vehicles, and Robotaxis are compliant. The standard clarifies that remote monitoring can only assist post-incident analysis, not rely on remote human intervention for road risk disposal (chǔ zhì - handling) . All failures, obstacles, and extreme conditions must be independently controlled by onboard systems. L4 vehicles in closed scenarios (e.g., tourist areas, ports, mines) may waive driver monitoring equipment but must include complete automatic failure handling modules.

3. Minimum Risk Mitigation (MRM): Full-scenario safety fallback

MRM represents the core safety fallback clause, establishing graded response procedures for three risk scenarios: system failures, ODD exceedance, and driver refusal to take over. Aggressive maneuvers like emergency braking or sudden lane changes that risk secondary accidents are prohibited.

Warning degradation phase: Limit maximum speed, prohibit automatic lane changes/overtaking, maintain continuous visual/audible system degradation alerts;

Emergency response phase: Smooth linear deceleration, activate hazard warning lights, gradually pull over to emergency lanes/roadsides while ensuring surrounding vehicle safety;

Parking lock phase: After complete vehicle stop, lock autonomous driving functions. High-level autonomous driving can only be reactivated after driver restarts the powertrain, preventing repeated functional activation risks under failure conditions.

4. Mandatory safety documentation and full-lifecycle corporate management

This standard innovatively introduces the internationally recognized Safety Case documentation system. All models applying for L3/L4 qualifications must submit complete structured safety documentation, including risk analysis reports, perception/decision hardware reliability verification data, tens of thousands of kilometers of simulation/real-vehicle testing records, human-machine interaction logic design, failure emergency plans, and OTA upgrade control procedures. Testing agencies verify each item, rejecting models with incomplete documentation from market launch (Source: China Industry News Network).

The standard also mandates automakers to establish complete safety management systems covering R&D, manufacturing, after-sales, and remote upgrades. Dedicated autonomous driving data recorders must fully document system activation, fault reports, takeover requests, and collision data, with retention periods ≥3 years for accident tracing and regulatory review. Strict OTA upgrade controls require full-scenario safety verification and regulatory reporting before deploying major algorithm updates, prohibiting untested mass deployment of unstable autonomous driving programs.

5. Three-tier closed-loop inspection system to ensure standard enforceability

To prevent standards from remaining theoretical, the standard includes a complete testing/acceptance mechanism forming a three-tier verification loop:

Tier 1: Verify corporate safety assurance systems, test sites, and simulation testing capabilities;

Tier 2: Comprehensively review corporate safety documentation and risk assessment materials;

Tier 3: Complete full-condition performance verification through simulation, closed-track testing, and open-road real-vehicle trials.

Only models passing all three inspection tiers can obtain L3/L4 type approval, raising industry access thresholds from the source.

This standard continues China's overall regulatory approach of "closed-scene prioritization, gradual open-road expansion," implementing tiered/categorized management for different automation levels and operational scenarios while avoiding one-size-fits-all controls to balance safety baselines and industrial innovation space.

From the automation level perspective, L3 targets consumer mass production as a transitional form for highway closed roads, with regulatory focus on human-machine interaction and driver takeover supervision. L4 focuses on commercial specialized vehicles covering closed parks, trunk logistics, and urban mobility, with regulatory emphasis on system autonomous risk handling, redundant hardware configuration, and safety assurance for unmanned regular operations.

From application scenarios, the standard relaxes some human-machine interaction constraints for closed/semi-closed scenarios (ports, mines, industrial parks, sightseeing vehicles), allowing compliant deployment of models without steering wheels or DMS monitoring. Highway trunk logistics emphasizes long-distance perception, platooning, and braking redundancy indicators. Urban open-road Robotaxi and buses face the strictest perception redundancy, extreme weather handling, and data retention requirements. L3 passenger vehicles are only permitted on highway sections, excluding city-wide conditional autonomous driving.

The tiered regulatory logic matches different scenario risk levels, prioritizing commercialization in controllable, simple-condition scenarios while setting higher safety thresholds for complex urban environments with mixed human-vehicle traffic. This facilitates phased, region-specific L4 deployment, achieving policy-technology-commercialization synergy.

The July 2027 implementation deadline will become an industry watershed, reconstructing development logic across four dimensions: automakers, upstream components, mobility operations, and insurance/legal support. While raising short-term compliance costs, it will accelerate long-term industry consolidation and drive high-quality, large-scale development.

In the short term, cost differentiation between leading and small automakers will intensify. Meeting requirements for perception redundancy, DMS monitoring, safety documentation, and simulation testing facilities demands increased R&D, testing, and hardware procurement investments. Small automakers lacking technical reserves and financial strength will struggle to comply, gradually exiting the L3/L4 track (sài dào - arena) and accelerating industry reshuffling. Marketing models relying on simplified hardware and exaggerated autonomous capabilities will become obsolete, as all promoted functions must align with national standard testing indicators while marketing regulations tighten simultaneously.

Long-term, unified national standards reduce cross-regional expansion costs for leading automakers, eliminating the need to adapt to differing local pilot rules and enabling nationwide unified launches/operations. R&D focus shifts from extreme function demonstration to system safety redundancy, extreme-condition algorithm refinement, and full-lifecycle risk control. Industry competition transforms from "parameter racing" to comprehensive evaluation of safety, reliability, and commercial operation capabilities. Safety documentation and accident data retention mechanisms help automakers establish complete fault iteration databases, accelerating data closed-loop efficiency.

For upstream component supply chains, perception, computing power, and monitoring hardware see incremental demand. The standard's performance mandates for perception systems establish multi-sensor fusion as the mainstream technical route, making single pure-vision solutions difficult to pass rain/fog/night attenuation tests. Market demand for 77GHz millimeter-wave radars, LiDAR, high-specification cameras, and vehicle-grade computing chips continues to grow. The lane-level lateral error requirement ≤0.5 meters for positioning modules drives growth in high-definition mapping and IMU inertial navigation supporting industries.

Standardized component testing indicators eliminate the need for suppliers to customize differentiation (chā yì huà - differentiated) adaptation solutions for different automakers, enabling scaled production to reduce hardware costs. Post-2027, high-level autonomous driving perception hardware prices will enter a downward trajectory, laying cost foundations for mid-to-long-term popularization. Supporting safety hardware like data recorders and onboard storage will become standard L3/L4 equipment, creating new market segments.

For operational enterprises like Robotaxi, autonomous heavy trucks, and driverless buses, the mandatory standard provides compliance basis for unmanned regular operations. Previous unmanned operations in multiple pilot regions lacked upper-level standard support, resulting in cumbersome approval processes. After standard implementation, L4 operational vehicles meeting ODD, MRM, and redundant perception requirements can streamline road approvals, accelerating commercial replication in ports, mines, and highway trunk logistics.

Clear accident liability divisions also resolve operational enterprises' biggest business pain point. Previously ambiguous autonomous driving accident liability led to insurance underwriting difficulties and high premiums. With the standard defining system/human responsibility boundaries, insurers can precisely calculate risks, accelerating deployment of dedicated autonomous driving insurance products and significantly reducing operational enterprises' claim costs—unblocking a critical path to commercial profitability.

As a technical foundation, the standard will drive matching (pèi tào - supporting) revisions to the Road Traffic Safety Law, clarifying legal subjects for autonomous vehicles on roads. Judicial authorities can determine primary/secondary accident liabilities based on standard clauses, forming standardized precedents. The insurance industry will establish tiered, scenario-based premium systems using standard risk indicators, with differentiated underwriting for L3 passenger vehicles and L4 commercial vehicles—resolving industry pain points of high-level autonomous driving insurance difficulties. Meanwhile, data retention mechanisms provide complete objective evidence for traffic accident forensics, significantly reducing dispute resolution cycles.

With just over a year remaining until formal implementation, some automotive companies have yet to establish comprehensive simulation testing platforms and safety documentation management systems, and their perceptual hardware redundancy designs do not meet the standards. To address this issue, the standard sets up a phased transition period: new models submitted for approval from July 2027 must comply mandatorily, while existing models that have already obtained qualifications will be given an additional year for rectification, providing small and medium-sized enterprises with sufficient time for technological iteration and production line upgrades. Companies can reduce individual investment costs by sharing resources through third-party testing institutions and joint simulation testing platforms.

The national standard requires coverage of a vast array of long-tail scenarios, including heavy rain, dense fog, tunnel lighting transitions, and sudden pedestrian crossings, making full-condition real-vehicle testing extremely costly. The industry can rely on national-level intelligent connected vehicle testing demonstration zones to jointly build and share extreme weather simulation facilities and million-kilometer simulation databases. Industry associations can take the lead in establishing joint testing platforms to share testing and validation costs and accelerate algorithm iteration.

In terms of supporting infrastructure, L3 highway autonomous driving and urban L4 operations rely on clear road markings and vehicle-road coordination equipment. However, some domestic highways and urban roads have aging markings and lack vehicle-to-everything (V2X) infrastructure, restricting the expansion of system design and operational domains. Subsequently, transportation and industry and information technology departments will simultaneously advance road intelligence upgrades, forming a coordinated pace of integrated upgrades for vehicles, roads, and standards.

The implementation of this mandatory national standard officially marks China's autonomous driving industry moving away from the pilot cultivation phase and entering a new stage of large-scale development characterized by 'safety first, compliance priority, and scenario-based implementation.' Three clear trends will emerge in the industry's development.

First, the implementation pace will strictly follow the path of 'closed operations first, followed by highway commercial vehicles, then urban mobility, and finally full-domain L4 for home use.' L4 operations in ports and mining areas currently have a complete compliance foundation and will rapidly scale up after 2027; autonomous driving heavy trucks for trunk logistics are expected to become widespread in 3-5 years; urban Robotaxi and driverless buses will gradually achieve full-domain safety officer-free operations in about 5 years; full-scenario urban L4 for ordinary consumers, constrained by complex urban scenarios and institutional support, is expected to be limitedly liberalized around 2035.

Second, the industry will form a positive cycle of 'policy support, technological iteration, and commercial feedback.' The national standard sets clear compliance boundaries, and policies will gradually relax scenario restrictions; companies will rely on unified standards for large-scale mass production and operations, accumulating real-world driving data to continuously optimize algorithms; commercial operations will generate stable cash flows, continuously supporting hardware R&D and safety technology upgrades. These three elements will mutually support each other, steadily expanding the operational boundaries of autonomous driving.

Third, China's domestic standard system will continue to be exported internationally, promoting the globalization of the autonomous driving industry. With a comprehensive mandatory safety standard framework, domestic vehicle manufacturers, autonomous driving solution providers, and component companies will have complete compliance endorsements for overseas expansion. Mutual recognition of Chinese and foreign standards will continue to advance, enabling Chinese autonomous driving technologies, vehicle products, and operational models to simultaneously enter global markets.

For all industry participants, this standard serves as the institutional foundation for the full-scale and commercialized implementation of autonomous driving. Vehicle manufacturers, component suppliers, and operation service providers should seize the one-year transition window to address safety technology gaps in line with the national standard and improve full-lifecycle safety management systems. Regulatory, insurance, transportation, and other supporting institutions should simultaneously advance the coordinated upgrades of laws, regulations, financial products, and road infrastructure.

Guided by unified standards, L3 and L4 autonomous driving will follow a scenario-based and region-specific progressive implementation path, steadily achieving the transformation from technological concepts to mature production factors, and propelling the autonomous driving industry toward a new stage of high-quality and sustainable development.