Either way, we are one step closer to fully autonomous driving.
Editor’s note: This article is from the micro-channel public number “car cloud” (ID: cheyunwang), Author: Yu Xiao.
Xiaopeng NGP officially pushed NGP (Navigation Guided Pilot) high-speed autonomous navigation driving technology through OTA, which is similar to the automatic navigation assisted driving function of Tesla NoA and NIO NOP, which is also currently available in China The third company has the ability to actively change lanes and enter and exit ramps in addition to autonomous cruise. In addition, the traditional car company Great Wall WEY also announced that it will launch a similar function of the Great Wall version of NOH in March. This automatic navigation assisted driving function is also regarded as one of the important signs of passing the current mainstream L2.
Tesla, the first to launch similar functions, launched a beta version of the autopilot system FSD Beta to a small number of users last year, showing us a lot of capabilities that have not been pushed in large numbers. As Tesla’s competitors began to follow up one after another, FSD Beta went through a few months of silent evolution.
Tesla’s NOA is secretly doing Beta testing
Since FSD Beta is still only experienced by a few car owners with close to 1,000 people, and it is mainly carried out in North America, we can only witness the capabilities of FSD through their experience videos. At present, FSD Beta has been upgraded to the latest version 2020.48.35.7 via OTA. As soon as it appeared, its current L2 autopilot capabilities and hard-core visual style have attracted attention every time it is upgraded.
Let’s first briefly enumerate some of the things that FSD Beta can do that we are currently hard to see in China:
– There is a vehicle in front of you parking at a side position, the system will slowly slow down and wait;
– Avoid pile buckets, avoid parking vehicles that are occupied by the roadside, and avoid vehicles occupying your own lane from the opposite lane;
– When encountering a vehicle that is approaching and too close in the adjacent lane behind the side, it will take the initiative to slightly dodge to the other side in the lane;
– Even if there is no obvious sign to indicate a closed lane in front of the construction, you can automatically change to another lane and change back after the construction area is over;
– The city will also change lanes about 1 mile in advance to prepare for a turn, and even if there is no clear lane line, it will pull over in advance;
– If there are two turning lanes when turning, turning a small turn on the inside and a big turn on the outside, it will not affect the vehicles in the adjacent lanes;
– Except for automatic stop at red light, automatic stop/start at stop line, automatic observation of intersections and turn (and prompt on the screen)Show) and so on through the action of the intersection, facing pedestrians who run red lights, turning vehicles that violate the rules and not to be courteous, etc. can also be courteous;
– In a single lane, tentatively turn to the left to observe the road conditions of the oncoming vehicle and overtake quickly;
– The mountain road does not rely on the lane or the vehicle ahead to drive autonomously. Even when there are no street lights at night, you can correctly choose irregular intersections, flexibly avoid vehicles parked on both sides of the road during driving, and stop when passing by After the vehicle moves to the right automatically;
– When encountering a speed bump, although there is no display on the screen, the vehicle will significantly reduce the speed to pass, but the obvious bumps on the road itself cannot slow down like other vehicles around.
These are some very daily scenes that rely on humanized operations. When Tesla FSD handles various evasive actions, steering, acceleration and deceleration, they are all smoother, and the speed is not slow at all, basically with our own The driving behavior is highly consistent, and is hardly affected by some rain, night and other weather. Moreover, in the United States, there are very clear orders and rules for crossing intersections. Sometimes we may not observe them in place. We don’t observe the vehicles in other directions carefully and don’t see the stop signs, while the NOA “sees” the other intersections for the owner. The vehicle in the direction is courteous.
In addition, in various scenarios, the left side of the central control screen will display the scene observed by the camera in real time, and the display area accounts for most of the display area, including all vehicles moving forward and backward, vehicles and pedestrians parked on the roadside, and trash cans, Pile buckets, road shoulders and other roadside objects that can be detected, and can be automatically or manually zoomed in/out, allowing you to synchronize its perspective.
So, from our current perspective, the overall strength of FSD Beta is far better than everyone else. But the reason it is a Beta version, there must be a problem with the test version, it is certainly not perfect. For example, at intersections where there are no traffic lights or even stop signs, the system will be overly cautious and will not dare to turn or pass, or even get stuck at the intersection and wait for a few minutes due to dense traffic; another example is for pedestrians or bicycles that violate the rules at the intersection. , The system is often very careful to slow down or stop waiting.
It is naturally necessary to consider from a safety perspective, but it may still have some impact on other traffic participants on the road.
Two ideas, two games
The powerful capabilities of FSD Beta can’t help but make many people think differently about Tesla’s pure vision + algorithm. This actually reflects the difference between Tesla and other car companies in terms of autonomous driving. In China, everyone generally adopts another way of thinking. For example, independent brands and new forces have gradually begun to adopt a comprehensive solution of camera + high-precision map + lidar.
TesPulling is obviously still a pure visual solution. For example, Model 3 now has a front-viewing trinocular, 4 side-viewing cameras, and a total of 8 rear-viewing cameras. The benefits of this are obvious. The cost of vehicle autopilot hardware can be controlled lower, and with Tesla’s huge database, accurate algorithms, and powerful chips, it can have strong capabilities. Musk has also been more than I once expressed my dislike of lidar.
Currently, FSD Beta has all the very powerful abilities we mentioned above. One of the reasons why it only relies on pure visual solutions + software algorithms is that Tesla has always used real road conditions for training. Nothing is more effective than practicing in a real environment. This is almost all other autonomous driving research and development companies. They are all incomparable. The reason is simple. No matter how realistic the simulation is, even if the test vehicle is directly driven on the open road, it still belongs to the test situation, which is always different from the reality.
During daily driving, we may encounter various unpredictable emergencies, such as sudden pedestrians, sudden explosions of fire hydrants, and sudden dumping of goods on the side of a truck, just like the movie “Bank Gang” When imitating the Italian Job here, the sudden appearance of the bicycle disrupted the whole plan. Although Tesla’s FSD, including ourselves, cannot encounter all unexpected situations, the experience and processing capabilities gained through these real cases and the unexpected situations that have been encountered are unmatched by others. This is one of the reasons why even though FSD is not a panacea, it is almost the best performer.
However, because it is only a Beta test version, domestic car owners cannot experience it yet, so the current domestic NoA does not represent Tesla’s strongest ability. However, the news of the Q4 conference call is expected to push FSD in China this year, and that the potential of FSD is still fully developed and fully supports fully automated driving in terms of computing power.
On the other hand, the current focus of Tesla’s Autopilot researchers, road testing, data analysis and other research and development forces is still in North America, and there have not been targeted tests for the more characteristic and complex conditions of domestic roads, like The aforementioned scenarios of poor FSD performance are very common on domestic roads, so the level that Tesla FSD can eventually reach in China is still to be developed.
On the other hand, NIO’s latest ET7, Xiaopeng’s new car P5, and BAIC’s new car HBT, Jihu, have all announced that they will be equipped with Lidar, which is the integrated camera + HD map + Lidar solution we mentioned above.
The most restrictive factor for this more advanced sensor solution was cost. However, as some car companies have begun to deploy lidar on mass-produced vehicles this year, this year is likely to be the first year of lidar on the domestic market. This also means that the cost of lidar has gradually dropped to an acceptable range. And for MuskThe pure visual solution is bad news, and its cost advantage is not so obvious.
With the blessing of lidar, in theory, as long as the algorithm and logic keep up, the effect is more perfect than pure vision. More advanced sensors can also make up for the lack of software algorithms to a large extent, or reduce the algorithm. Requirements and standards. However, this solution has high requirements for vehicle computing power and energy consumption control, which may affect the endurance of an electric vehicle, and even adjust the battery layout and three-electric system from the beginning of the research and development. These are also equipped Issues to consider after lidar.
But at least for now, Tesla still believes in its cameras. Not long ago, Musk also replied to netizens on Twitter that they are gradually covering all the neural network algorithms on the surrounding images of 8 cameras to achieve autonomous driving capabilities far exceeding human capabilities. Although it is very complicated to integrate the entire labeled and inferred data into the surround vision, the advantage is that there is no need to replace any hardware. In other words, this may be another capability upgrade that can be achieved through OTA, and Tesla’s vision solution is still further rapidly evolving.
Of course, this does not mean that Musk will definitely go to “black” in pure vision. If the conditions for carrying other hardware are mature, the cost and threshold can meet his universal and popular autopilot The request, then why not do it.
Summary of Che Yun
We have seen the powerful capabilities of FSD on Tesla, and we have also seen future potential on our own brands. But no matter which option, we are one step closer to fully autonomous driving. Musk once again said a month ago that he would have L5 fully autonomous driving capabilities this year, and he also made a $10,000 bet with people. Some are awesome and some are bragging, but we still hope it can be realized from the bottom of our hearts, we will wait and see.