Recently, an incident involving the disengagement of an advanced driver-assistance system (ADAS) on a highway has drawn attention. According to a report by China Business News on October 5, a vehicle was traveling at approximately 120 km/h with navigation-assisted driving activated. When the vehicle was about 10 meters away from a large truck ahead, the system disengaged the assisted driving function and prompted the driver to take over, resulting in a rear-end collision.
A speed of 120 km/h means the vehicle travels approximately 33 meters per second; a 10-meter distance only allows the vehicle to continue moving for about 0.3 seconds. Even with instantaneous maximum braking, the pure braking distance at 120 km/h typically requires 50 to 70 meters; adding the driver's reaction time, the total stopping distance will far exceed this figure. For the driver, the process from noticing the system prompt, confirming the current traffic conditions, to applying the brakes or steering, inherently takes time.
Studies have shown that drivers take an average of 6 to 8 seconds to regain situational awareness. This actually raises a frequently discussed question: while intelligent driving systems can disengage, does the timing of the disengagement leave sufficient room for the driver to take over?
01. Intelligent Driving Disengagement Does Not Mean the Driver Has Completed the Takeover
The disengagement of an ADAS typically has corresponding trigger conditions, such as the current road environment exceeding the system's Operational Design Domain (ODD), anomalies in key sensors or the system, or the system determining that it can no longer continue the current driving task. However, the system recognizing its inability to continue operating and the driver completing the takeover are not accomplished in the same instant.
Assume a vehicle is traveling at high speed, and the system determines that the current scenario exceeds its processing capabilities. If the vehicle is already very close to a risk ahead at this moment, even if the system issues an immediate prompt, the driver still needs to recognize the system prompt, assess the current traffic environment, and then take actions such as braking or steering. The less time left for the driver, the higher the difficulty of completing safe operations.
It is also necessary to supplement the basic responsibility relationship in Level 2 (L2) assisted driving. During the operation of L2 combined driver assistance, the driver still bears the dynamic driving task, while the system provides lateral and longitudinal driving assistance. Therefore, system disengagement does not mean the driver had no driving responsibility beforehand, nor should it be understood as the driving task being handed over to the driver for the first time only after the system disengages.
However, from a system design perspective, the timing of disengagement remains crucial. System disengagement is merely the state where the assisted function stops working, whereas the driver completing the takeover requires going through an actual operational process. The driver needs to confirm the vehicle's current speed, lane position, distance to the preceding vehicle, and the status of surrounding traffic participants, and then decide whether to brake, steer, or continue driving. Therefore, the disengagement mechanism must consider not only whether the system can continue driving but also whether the driver has sufficient time to complete safe operations when the disengagement is proposed.
Especially in high-speed scenarios, if the system can issue a prompt before the risk becomes imminent, the driver has more time to observe the road and take action; if the vehicle disengages only when it is already close to a dangerous target, the time available for the driver to take effective operations will be significantly reduced.
02. Why Must the Takeover Request of an L3 System Last for at Least 10 Seconds?
On July 30, 2026, the "Intelligent Connected Vehicles - Safety Requirements for Automated Driving Systems" (GB 44721-2026) was officially released, planned to be implemented on July 1, 2027. This standard applies to Category M and Category N vehicles equipped with L3 and L4 automated driving systems. In the interpretation of the standard, the Ministry of Industry and Information Technology (MIIT) explicitly stated that the activation and disengagement processes of the automated driving system should be safe; for L3 automated driving systems, it is also required to have a driver takeover capability monitoring function. One of the requirements directly related to takeover is that after the L3 system issues an intervention request, the time from the issuance of the request to the termination of the request due to the execution of a minimum risk strategy should be no less than 10 seconds.
Image source: Screenshot of the standard
It does not mean that all intelligent driving systems must have a 10-second countdown before disengagement, nor does it mean the system must turn off assisted driving 10 seconds in advance. This requirement targets the human-machine handover process after the L3 automated driving system issues an intervention request. There is an important distinction between L2 and L3 here.
During the operation of L2 combined driver assistance, the driver always bears the dynamic driving task; L3 automated driving, on the other hand, has the system execute the dynamic driving task when the operational design conditions are met. When the L3 system is about to be unable to continue operating, it needs to issue an intervention request to the driver (fallback-ready user) to have the driver resume the driving task.
Therefore, the safety issues of L3 are no longer just about when the system disengages, but also include when to issue the intervention request, whether the driver has the takeover capability, and what the vehicle should do if the driver fails to complete the takeover. GB 44721-2026 precisely places these links within a set of safety mechanisms.
The standard requires the L3 system to have driver takeover capability monitoring and sets requirements for the intervention request and the subsequent minimum risk strategy. If the driver does not complete the takeover, the system cannot simply stop working but must continue to adopt safety strategies according to corresponding conditions. Therefore, the 10 seconds in the standard, more accurately, is the lower time limit set for the duration of the L3 system's intervention request, rather than stipulating a unified countdown for intelligent driving disengagement.
03. The Size of the Takeover Window Depends on When the System Identifies the Problem
For vehicles traveling at high speeds, the takeover time is closely related to the vehicle speed, road environment, and the location where the risk appears. Calculated at 120 km/h, the driving distance corresponding to 10 seconds exceeds 330 meters. This does not mean the standard requires the system to detect the danger 330 meters in advance, let alone be understood as all L3 scenarios must issue a takeover request at a distance of 330 meters from the dangerous target. The later the system realizes it cannot continue driving, the less time is left for the driver to handle the current situation. Therefore, the system needs to identify its capability boundaries as early as possible while there is still sufficient safety margin, and issue an effective takeover request in a timely manner.
For the L3 system, judging whether the driver has the conditions for takeover is equally important. GB 44721-2026 has already incorporated driver takeover capability monitoring into the requirements, indicating that the system cannot simply assume the driver can take over the vehicle immediately at any time. This is also where L3 automated driving is more complex than traditional assisted driving; the system not only needs to know whether it can continue driving but also needs to know when to hand over the driving task to the human, and whether the driver has the conditions to safely take over the vehicle.
04. Concluding Remarks
For this L2 assisted driving incident, the not-yet-implemented GB 44721-2026, which targets L3/L4 automated driving systems, cannot be used to judge whether it meets the 10-second requirement. The two belong to different levels of automated driving and correspond to different safety requirements. In addition, from a system design perspective, disengagement, issuing a takeover request, and the driver completing the takeover should inherently be a continuous human-machine handover process.
The boundary of intelligent driving capabilities is not just about when the system can no longer continue driving, but also includes whether the system can identify problems before reaching the boundary and leave sufficient time and safety margin for the driver.