On October 19, Xiaomi Auto announced the OTA update for Xiaomi HyperOS 1.18, focusing on optimizing the urban assisted driving experience for the SU7 and YU7, including improvements in acceleration, deceleration, lane changing, obstacle avoidance, and intersection navigation. One specific optimization for Highway NOA explicitly mentioned enhancing the aggressiveness of overtaking.
Why does the autonomous driving system need to specifically optimize overtaking aggressiveness?
For Highway NOA, overtaking does not mean immediately executing a lane change just because a gap is detected in the adjacent lane. The system also needs to evaluate how much traffic efficiency would be lost by continuing to follow the lead vehicle, how much benefit overtaking would bring, and whether the entire maneuver can be completed under safety constraints.
01. Why Does the Autonomous Driving System Prefer to Wait for Some Overtaking Opportunities?
Assume the vehicle is driving on a highway, with a slower vehicle ahead traveling below the ego vehicle's target speed, and a temporary available gap in the left lane. A human driver might quickly decide to overtake, but the autonomous driving system needs to compare the expected outcomes of two behaviors: continuing to follow the lead vehicle and executing an overtaking maneuver.
If it continues to follow, the ego vehicle's speed may remain constrained by the lead vehicle; if it chooses to overtake, it needs to complete a lateral lane change, adjust longitudinal speed, and re-establish an appropriate car-following relationship after entering the target lane.
Both behaviors may satisfy current safety requirements, but their impacts on subsequent driving states are not the same. Therefore, what the system needs to evaluate is not simply the speed of the lead vehicle, but whether this overtaking maneuver can bring sufficient traffic benefits. If the speed difference between the ego vehicle and the lead vehicle is very small, completing the overtake may not necessarily improve the subsequent driving state significantly.
If another slow vehicle is encountered shortly after overtaking, the vehicle may have to decelerate again, limiting the benefits brought by the previous lane change. In this case, even if a safe gap exists in the target lane, it does not mean that overtaking immediately is the better choice.
Conversely, if the lead vehicle continuously restricts the ego vehicle's speed and the target lane can provide a more stable driving space, timely overtaking may reduce the time spent following at low speeds.
Here, it is also necessary to distinguish between speed benefits and time benefits.
An increase in vehicle speed after overtaking does not mean that the trip time will necessarily be shortened by the same proportion. The actual benefit also depends on the duration of the overtaking maneuver, how long a higher speed can be maintained after completing the action, and the traffic conditions on the subsequent road. From the perspective of algorithm design, this can be understood as evaluating candidate behaviors. The system needs to compare the expected outcomes of different choices, such as continuing to follow, executing an overtake, and waiting temporarily, and select the more appropriate behavior under the premise of satisfying safety constraints.
If the strategy is too conservative, the system may miss overtaking opportunities that already meet the execution conditions; if it overemphasizes traffic efficiency, it may change lanes frequently without gaining corresponding speed or time benefits. Therefore, enhancing overtaking aggressiveness should not be simply understood as lowering the threshold for lane changes, but should focus on improving the identification and evaluation of overtaking opportunities, enabling the system to execute valuable overtakes more timely while reducing maneuvers with limited benefits.
02. Why Can't the Target Lane Be Judged Solely by Front and Rear Distances?
Even if the system determines that overtaking is beneficial, it still needs to judge whether the target lane has sufficient safety margins.
There is an easily overlooked detail here: the distance between vehicles is only the current state, and the relative speed between the two parties determines how this distance will change subsequently.
Assume the ego vehicle is preparing to change lanes to the left, and a vehicle behind in the target lane is 60 meters away from the ego vehicle, but the rear vehicle is 10 meters/second faster than the ego vehicle. If we temporarily assume both maintain this speed difference, the distance between the rear vehicle and the ego vehicle will shorten by 10 meters per second, shrinking from 60 meters to 30 meters after 3 seconds.
This example only describes simplified longitudinal relative motion and cannot be directly used to judge whether 30 meters is safe. The actual risk also depends on the absolute speeds of the ego vehicle and the rear vehicle, whether either party is accelerating or decelerating, the lateral position to which the ego vehicle moves, and whether the future trajectories of the two vehicles might conflict.
However, this also illustrates that the system cannot merely judge whether there is currently a sufficiently large gap; it also needs to determine whether this gap can maintain an acceptable safety margin throughout the entire lane-changing process. This involves target tracking, motion prediction, and safety gap evaluation.
The system needs to predict the possible motion states of surrounding vehicles over a future period based on their positions, speeds, and motion trends. Then, combined with the ego vehicle's lane-changing trajectory and estimated execution time, it evaluates whether the target lane can maintain a sufficient safety margin throughout the lane-changing process. Here, the lane-changing time itself is not a fixed value independent of the scenario. Vehicle speed, lateral motion planning, steering response, and target trajectory all affect the duration of the maneuver. The longer the lane change takes, the more the system needs to consider the motion changes of surrounding vehicles during this period.
Therefore, the safety gap cannot be determined solely by a fixed distance. The same distance may correspond to completely different risks under different speed differences and traffic conditions. Even if the safety gap meets the requirements, it only indicates that overtaking may have the conditions for execution, not that it is definitely worth executing. The system still needs to make a comprehensive judgment combining the speed loss caused by the lead vehicle, the traffic state after completing the overtake, and the execution cost of the entire maneuver. Safety gap evaluation primarily answers whether current conditions for overtaking are met, while benefit evaluation answers whether overtaking now is more appropriate.
The former focuses on whether the maneuver can be executed safely, while the latter focuses on whether executing this maneuver has sufficient value. The two are interrelated but cannot replace each other.
03. Why Might a Beneficial Overtake Be Temporarily Unexecutable?
The overtaking decision actually faces another issue: just because the behavioral level deems it worth executing does not mean the trajectory planning level can necessarily generate an appropriate execution plan. Assume there is an available gap in the target lane, and the system also determines that overtaking can improve traffic efficiency. However, if the current vehicle speed is high and the available space is limited, the vehicle needs to coordinate steering and longitudinal speed changes while satisfying lateral motion constraints.
If the lateral movement lasts too long, the vehicle may maintain prolonged interaction with surrounding traffic participants during the lane change; if the lateral action is too abrupt, it may bring significant lateral acceleration, affecting passenger comfort. Therefore, trajectory planning cannot merely pursue curve smoothness; it also needs to consider vehicle dynamics capabilities, lateral acceleration, longitudinal speed, and the relationship between the trajectory and the predicted motions of surrounding traffic participants. This is also the reason why coordination is needed between behavioral decision-making and trajectory planning.
Behavioral decision-making is responsible for judging whether to overtake and when to initiate it. Trajectory planning is responsible for generating specific paths that satisfy vehicle motion constraints and safety requirements. The control system then executes steering and speed adjustments based on the actual state of the vehicle.
If the planning results cannot meet current conditions, the system cannot forcibly execute the original action simply because it has already decided to overtake. It needs to re-evaluate feasible behavioral plans and adjust the trajectory or timing when safety conditions permit. Even after a lane change has been initiated, the judgment cannot stop. The vehicle behind in the target lane might accelerate, the vehicle ahead might decelerate, and the originally feasible gap might change. The system needs to continuously update the motion states of surrounding vehicles and judge whether the current trajectory still meets safety requirements.
This does not mean that overtaking must be abandoned as long as surrounding vehicles change; rather, subsequent actions need to be re-evaluated based on the changed traffic state and the vehicle's position. Continuing to complete the maneuver, adjusting the trajectory, or taking other safety measures should all be based on the actually feasible conditions at that time. Thus, overtaking aggressiveness cannot be evaluated in isolation from trajectory quality and execution capability. Initiating a lane change earlier does not necessarily mean higher efficiency; completing the overtake faster does not necessarily mean the entire maneuver is more reasonable.
04. Final Thoughts
Evaluating the overtaking capability of an autonomous driving system cannot rely solely on counting how many times the vehicle has completed an overtake. It is more important to observe whether the system can identify truly valuable opportunities, whether it can maintain sufficient safety margins in dynamic traffic environments, and whether the decision results can be executed smoothly and reliably by the vehicle. Progress in overtaking capability is not about making the vehicle more eager to pass the lead vehicle, but about ensuring that every overtake has sufficient justification, along with matching safety conditions and execution plans.