Tesla's assisted driving has always been one of the most topical hot spots in the car circle and even the technology circle, and now it has attracted a wave of widespread attention about Tesla FSD in foreign countries.
And this wave is mainly aimed at "scaring" Tesla FSD, and the thing is this:
A billionaire named Dan O'Dowd recently ran for the Senate in California, but one of the main campaign slogans was a stark opposition to the use of Tesla FSD on public roads.
To this end, the fierce opponent plans to invest $2 million to put a video about how Unreliable Tesla FSD is in 36 states in the United States to increase public opinion.
This is a bit outrageous...
After all, this is not as simple as a tech guru or KOL tweeting.
A billionaire running for state Senate (Dan O'Dowd's background is also the founder of a large software company) has put his opposition to Tesla FSD on a political slogan, and Tesla FSD is also extraordinary...
Whether or not the opponent ultimately campaigned, this is at least a planned large-scale attack on Tesla's FSD project.
But as a melon-eating masses, let's first look at this hundreds of millions of opponents.
Attacking Tesla FSD is not reliable, is it reliable?
In fact, Dan O'Dowd's main objection is a one-minute short video called "Unsafe at Any Speed" that "any speed is not safe."
Exclude the "ad" in which Dan O'Dowd introduces himself at the beginning of the film, and the part at the end that explains who paid for the film.
In the remaining fifty seconds of the short film, only one thing was simply and rudely edited. It is to take the past period of time, YOUTOBE about Tesla FSD in different versions of the wrong case, to make a collection.
The video is very detailed about the test version of the FSD and the timing.
The whole film has a total of 10 cases, we have counted them, let's take a look at them:
1. Lane line recognition error, green pile bucket not recognized (FSD Beta v10.10)
2. In the case of complex ambient light, the lane is not correctly identified and almost hits the pole. （FSD Beta v10.2）
3. The intersection lane line is identified incorrectly. （FSD Beta v10.9）
4. Suspected of being in a community road without lane lines, the system sends out a prompt for the driver to take over. （FSD Beta v10.6）
5. The road section of the construction road at night fails to identify the roadblock in time. （FSD Beta v10.2）
6. Through the intersection, it is suspected that the truck turned right when going straight. （FSD Beta v10.9）
7. Suspected FSD left cut lane overtaking, but was twisted by the driver (FSD Beta v10.9)
8, unprotected left turn, but there seems to be no problem ... （FSD Beta v10.9）
9, unprotected left turn, it seems that just did not quickly identify the correct route, was briefly misled by the roadside parking line. （FSD 2020.40.8.11）
10. Through the intersection without traffic lights, when the car comes on the left side, it is suspected that the brake is too strong. （FSD Beta v10.6.1）
Looking at the whole film, excluding the last three cases that obviously do not seem to be a big problem in the video, there are still 7 left.
Of the remaining 7 cases, the fourth one also had no emergency, only that the system prompted the driver to take over without lane lines.
In the 6th and 7th cases:
In the sixth case, through the tesla vehicle instruments, the system had recognized the truck, but it was estimated that the deceleration was slower, and the driver actively stepped on the brakes.
In the 7th case, it seems that a convoy was lined up in front, and the FSD was ready to cut a lane to the left, but was taken over by the driver, and the only thing that seemed to be blamed was that the FSD steered a little harder.
In the end, there are 4 remaining cases, all of which appear in the identification of targets, mainly the identification of lane lines by FSD, followed by green pile barrels and roadblocks in construction sections, which are unusual traffic situation identification.
What do you say after watching this video? I feel that this video against Tesla FSD is slightly less careful.
Tesla FSD test version very early on small batch open test, foreign Twitter and YOUTOBE should be able to find a lot of FSD errors in the classic scene.
But less than half of the cases in this video that can clearly see FSD errors seem to be less than half, and they are all placed in the first half of the film, which gives people the feeling of paddling.
As for the FSD's failure to identify lane lines as well as barricades and green pile barrels.
In fact, first of all, we know that the current mechanism of FSD relies only on visual recognition, and there is no high-precision map involved in "cheating", and there is no lidar and millimeter-wave radar to play aids.
Although this does mean that in some complex road cases, the recognition of FSD may not be very good, such as no high-precision map, the judgment of the lane line completely needs to be completed by the camera itself, and secondly, if the visual neural network lacks targeted training on special-looking barricades and pile barrels, it is indeed impossible to effectively identify them.
But that doesn't mean FSD itself isn't good or unreliable, it's just the opposite.
The whole one-minute video we can see but the road environment is almost all complex community or urban surface roads.
Most of these are precisely the weaknesses faced by other assisted driving systems that require the participation of high-precision maps.
To put it this way, tesla FSD errors in the road without high-precision maps is not to prove that Tesla FSD is not good, but anyway, it proves that FSD can still show a certain degree of assisted driving ability in the absence of high-precision maps.
In addition, since the Tesla FSD Beta version itself is an unofficial version, it is also very normal to have errors, which does not mean that the FSD is not safe.
In Tesla's rules, users who can participate in the beta version of the FSD have a safety score of more than 98, and the safety score depends on the owner's daily habits of using FSD.
This means that you must not only be an old driver, but also have the concept of safe use of assisted driving, which is also a test of Tesla for users who have the use of assisted driving systems, and it has also raised a threshold in safety.
Musk has another flag, is the full version of the FSD landed this year?
Today' Tesla FSD Beta V10.11.2 is optimized at an almost weekly rate, with each update improving on some of the key issues that preceded it.
For example, in the FSD Beta V10.11.1 version, it is possible to more accurately predict the turns or parallels of other vehicles, thereby reducing unnecessary deceleration, improving the vehicle's understanding of the right of way at intersections, and improving the detection accuracy of pedestrians and bicycles.
Musk said a few days ago that Tesla's FSD test program currently has more than 100,000 drivers involved and will achieve full self-driving this year.
However, regarding the real landing time of FSD, Musk has been punched in the face many times.
At present, according to specific plans, Tesla has opened the FSD Beta to the Canadian market in March this year. Subsequently, it will be the first to land FSD in the European market, but it is not an L5 level autonomous driving capability, and it is still the FSD Beta version.
Tesla is expected to submit the FSD Beta to European regulators in the middle of this year, and the main goal should not be to promote the use, but to get more car owners to help run the data.
In fact, considering the characteristics of Tesla FSD, compared with the need to use high-precision maps and radar auxiliary driving solutions, Tesla FSD advantages are universal, on the one hand, it is a strong self-learning ability to the environment, to adapt to more road conditions, on the other hand, due to the great reduction of hardware costs, users can enjoy at a lower cost.
Compared with other solutions, FSD is a bit more than just the scope of self-driving machines, such as similar to general-purpose AI robots, more like strong artificial intelligence.
Finally, back to the most critical question, when will Tesla FSD enter the country?
This may have to wait longer, and entering the country may not only face problems in terms of policies and regulations.
For the relatively more complex road conditions in China, such as the traffic signs in Shenzhen and Shanghai may be different, which requires the visual neural network to be trained separately for different situations in the two places, and for Tesla, which is not yet prepared, FSD will undoubtedly face more problems in the real landing in China.