Changing lanes without driver confirmation will be the next smart car trend

There are many models with SAE L2 level of autonomous driving nowadays, and Tesla even had this function many years ago. But it must be said that the current SAE L2 autonomous driving is still based on single lane keeping + adaptive cruise + AEB emergency braking as the basic functions. If you want to change lanes, you still need to confirm manually.

Take Cadillac's Super Cruise super intelligent driving system as an example. When the driver changes lanes, he needs to turn on the turn signal manually. At this time, the light bar on the steering wheel turns blue. Only when the vehicle reaches the target lane and drives in the center, and waits for the vehicle to change lanes, the driver will turn on the turn signal manually.

The driver can take his hands off the steering wheel only when the steering wheel light bar turns green again.

The most typical example is the car accident that occurred in California on March 23 last year, in which a Model X crashed into a barrier and killed the driver. Although the report details show that the main person responsible for the accident is the owner, if the self-driving car can achieve automatic lane change, it can at least help the driver gain more reaction time.

1What is the difficulty in changing lanes?

It would be great if self-driving cars were smarter and could change lanes to avoid slow cars, just like we do in daily driving. However, relevant surveys show that only for common manual lane changes, traffic accidents caused by it account for 5%, and traffic delays are as high as 10%. If only for highway driving environments, the accident rate related to automatic lane changes will soar to 60%. Specifically, 75% of lane change accidents are caused by the driver's lack of perception of the surrounding environment when changing lanes.

In current research, most lane-changing decisions focus only on the free lane-changing behavior of vehicles on straight roads, but there are few studies on the forced lane-changing behavior on the off-ramp of highways, and few models can quantify the relationship between lane-changing intention and lane-changing success rate in this scenario. Data show that the time interval and the relative speed with the adjacent leading vehicle are crucial to whether to change lanes. However, the current autonomous driving methods do not consider the time interval and relative speed.

In public statistics of autonomous driving demonstration data, the number of lane changes is often promoted as an important reference indicator. It can be said that in the next trend in the field of autonomous driving vehicles, whoever finds a solution will seize the initiative.

2Leaders have emerged

Recently, according to foreign media reports, Tesla's Navigate on Autopilot system has been upgraded to address this problem, and will no longer require driver confirmation to automatically change lanes.

Tesla said in the announcement: In the Autopilot settings directory, the driver can press the custom Navigate on Autopilot button, which can display three additional settings: Enable before each trip, Lane change information required, and Lane change notification. If the driver selects "Do not require confirmation of lane change", the lane will be changed automatically.

It is still unknown how Tesla achieves this function, but it is definitely related to the new AI chip they announced not long ago. This chip is expected to be released in the next few months and will be used in Tesla S, X, 3 and other models. AI chips are also called AI accelerators or computing cards. As the name suggests, they are specially used to handle large amounts of calculations in artificial intelligence applications. AI chips can provide the training algorithms and high-performance computing power required by machine learning technology, which just meets the short-term large-scale computing needs required for automatic lane changing.

In addition, some models based on specific road sections also have automatic lane change functions. For example, the experimental center covered by V2X function has a strong network communication support, which greatly reduces the computing needs of the vehicle. In other words, if vehicles can communicate with each other, this problem will become much simpler.

Automatic lane changing is a basic driving requirement, but the current autonomous driving assistance system is not good enough. Now that single-lane autonomous driving has matured, I believe that the focus of the next stage will be automatic lane changing.

Of course, it doesn't mean that changing lanes frequently is a good thing. I believe everyone has been annoyed by uncivilized lane changing behaviors. So in addition to the feasibility and execution of lane changes, we also need to pay attention to the motivation for lane changes. For example, why do you need to change lanes? Is it because you need to change lanes to turn left at the intersection ahead, or because you dislike the slow speed of the car in front of you, or because there is a tanker in front of you and you have to get away from it quickly? In short, we must have a reasonable motivation for lane change.