Woofun AI reports that the Ministry of Industry and Information Technology (MIIT) has issued the draft 'Safety Requirements for Autonomous Driving Systems in Intelligent Connected Vehicles,' establishing the first mandatory national standard for autonomous driving in China. This regulatory framework explicitly targets Level 3 and Level 4 systems, moving beyond the vague definitions that previously allowed manufacturers to market systems as 'L2.999+' to avoid liability. The standard fundamentally alters the operational landscape by defining specific capabilities and safety thresholds that distinguish true autonomous driving from assisted driving, effectively ending the era of ambiguous marketing claims regarding system responsibility.

Structurally, the new regulation delineates a clear boundary between Level 3 and Level 4 autonomy based on design limitations and user intervention requirements. Level 3 systems are defined as having specific design constraints that still necessitate user intervention when the system reaches its operational limits. In contrast, Level 4 systems possess no such limitations and do not require user assistance, a distinction that theoretically permits the removal of traditional safety hardware such as steering wheels and seatbelts. Consequently, the restrictive protocols applied to Level 3 operations do not automatically extend to Level 4, creating a tiered regulatory environment where higher autonomy levels enjoy greater operational freedom provided they meet stricter safety criteria.

Regarding functional capabilities, the standard mandates that Level 3 systems must operate effectively in high-speed and urban expressway environments. If these systems are deployed in general urban driving scenarios, they must demonstrate the ability to maintain lane position, execute lane changes, and navigate intersections without human input. Level 4 systems are held to an even higher bar, requiring advanced capabilities such as obstacle avoidance during lane changes, executing U-turns, and performing reversing maneuvers. While these functions may appear familiar to users of current L2.9+++ assisted driving systems, the new standard introduces a critical legal distinction: the system must ensure the driver remains alert and focused, and the Autonomous Driving System (ADS) must not pose an unreasonable risk to users or other road users.

The legal weight of the term 'must' in this document transforms these capabilities from optional features into obligatory requirements. Within the designed operational design domain, Level 3 and above systems are required to achieve a 100% accuracy rate in identifying critical objects such as road markers and other vehicles.

Furthermore, the system must possess the cognitive ability to recognize special emergency vehicles, including police cars, fire trucks, and ambulances, and determine appropriate actions based on its own capabilities. This requirement eliminates the ambiguity that previously allowed manufacturers to claim partial responsibility, forcing a binary outcome where the system either meets the 100% identification standard or fails to qualify for the higher autonomy classification.

A more critical variable is the protocol for system disengagement and risk management, directly addressing the controversial scenario of a system disengaging 0.1 seconds before an accident. The new standard explicitly prohibits such late warnings, mandating that if the system cannot handle a situation, it must alert the driver at least 10 seconds in advance using sound, light, or tactile feedback. This 10-second window is significant; it allows a sprinter like Usain Bolt to cover 100 meters, highlighting the necessity of predictability for Level 3 systems. If the driver fails to respond within this 10-second interval, or if a serious malfunction prevents timely alerting, the vehicle must immediately implement the Minimum Risk Strategy (MRM).

The MRM protocol does not involve an immediate shutdown or function downgrade that could cause a sudden stop in traffic. Instead, the vehicle is required to slow down, change lanes if necessary, and come to a halt in a location that does not obstruct traffic flow. If all attempts to move to a safe location fail, the vehicle must stop safely in its current lane.

Additionally, Level 3 systems must continuously monitor driver attention, utilizing methods such as tightening seatbelts or emitting warning sounds to alert inattentive drivers. If the driver remains unresponsive, the system automatically enters the Minimum Risk State (MRC) and stops the vehicle, ensuring that the driver is protected even in the event of total inattention.

Appendix D of the standard introduces rigorous hardware redundancy requirements for the vehicle itself to ensure system resilience. In the event of an autonomous driving system failure, backup mechanisms must be immediately available. For instance, if a lidar sensor is damaged, the system must utilize two additional sensors as backups, or alternatively, employ millimeter-wave radars and cameras as supplements to maintain situational awareness. It is also essential to maintain multiple copies of the system software to prevent total system failure due to software corruption. Not all vehicles currently equipped with autonomous driving systems possess these redundant mechanisms, but their inclusion in the national standard accelerates the adoption of these critical safety technologies.

The standard also specifies detailed physical requirements for the lane-changing process, rejecting arbitrary cutting-in behaviors often seen in earlier iterations of intelligent driving. When changing lanes behind oncoming traffic, the autonomous driving system must ensure that the approaching vehicle does not decelerate by more than 3 m/s² and that there is at least 1 second of distance between the two vehicles. This requirement ensures that lane changes are executed smoothly without forcing other drivers to brake abruptly. When changing lanes to overtake, the vehicle must increase its speed and utilize redundant mechanisms to avoid interfering with other traffic, effectively codifying the behavior of a truly skilled human driver into algorithmic logic.

Woofun AI data shows that these physical constraints create a complex liability landscape where the standard alone cannot determine compensation if a responsible autonomous vehicle is involved in an accident caused by a distracted human driver. The regulation requires car manufacturers to establish comprehensive safety records that include information about the expected capabilities of the autonomous driving system, its operating logic, hardware configuration, and principles of operation. These records must cover not only road tests and internal tests but also the simulation software used and the models developed, all of which are subject to strict supervision. Inspectors will verify whether the manufacturer's statements, arguments, and evidence are complete, traceable, and reproducible.

A pivotal metric in this regulatory framework is the failure rate requirement, which must meet the ASIL-D (Automotive Safety Integrity Level D) standard. This standard dictates that the system must have no more than 10 failures per 1 billion hours of operation, a threshold that represents an extremely high level of reliability.

In addition to meeting this failure rate, car manufacturers must implement Safety Assurance Requirements (SMS), establish robust quality management systems, effectively identify risks, clearly define responsibilities, and conduct regular internal and external audits. These procedures ensure that the autonomous driving systems are not merely compliant on paper but are rigorously tested and monitored in practice.

The implementation of these regulations marks a shift from unregulated experimentation to a highly controlled environment where safety is the paramount concern. While the increased safety measures may result in autonomous driving systems becoming more conservative and less versatile than initially anticipated, this trade-off is essential for public acceptance and safety. The standard aims to reduce accident rates and make accidents easier to explain, thereby fostering trust in the technology. By enforcing these strict guidelines, the authorities intend to leverage the safety advantages of autonomous driving to reduce accident costs and encourage wider adoption through economies of scale.