● Unmanned driving: Very polite, first hit the turn signal.
This function is more powerful than expected, and it can be calculated in real time by camera and millimeter-wave radar (it doesn't depend on maps and can't be used on open roads). Model? Just like a toddler groping his way forward, he will try his best to take shortcuts and take a straight line, which will be faster than expected, and the radar will measure the distance accurately and make way for pedestrians in time.
● The turning speed is very fast, and passers-by didn't find anything wrong with the car.
When pedestrians approach a certain distance, they will stop. This process did not arouse the vigilance of passing girls.
However, this function is mainly aimed at large outdoor parking lots in the United States, and the function of directional development in China is still somewhat unaccustomed. First of all, in China, there are not many large regular parking lots for you to stop production every day. It is normal to crowd together in the community, and the pedestrian density is extremely high. Obviously, such an environment is difficult for the summoning function. Once you go in the wrong direction or walk into a dead end, you will immediately enter a "stupid" state, and you will never look back in this life.
After all, the forward-looking camera and radar are still much stronger, and safety comes first.
Therefore, the practical degree of this function in China must be discounted, and most of them are dazzling skills.
But this is automatic driving, NOA (automatic assisted navigation and driving), intelligent call and other functions. Step by step. Let's see Tesla gradually turn the impossible into possibility, from concept to reality, and even free his hands. This is more important.
Tesla mode, is it the strongest?
Uncertainty must be the strongest, but Tesla's model is indeed unique and has its own characteristics. Just like Apple, it is better than software, but it has the ability to drive both software and hardware.
Maybe you think Tesla's autopilot is very powerful because his hardware is very advanced? Actually, it's not. On the contrary, Tesla's hardware part can be described as "well known". The reason is that Tesla, contrary to the normal state in other countries, has completely disclosed its sensor configuration, and you can even copy it and make the same layout. This is precisely the greatness of Tesla, just as it announced a series of related patents earlier.
How does Tesla insist on doing this? Tesla's models have been equipped with complete hardware supporting autonomous driving, but in actual experience, Tesla believes that autonomous driving ability cannot be achieved overnight, and it should be gradually improved and opened up by continuously optimizing the software system.
Therefore, the software or algorithm is the core of Tesla's entire autopilot system. Its hardware part is nothing more than the combination of camera, millimeter wave radar and ultrasonic radar, and no tall (expensive) lidar is used.
As for the main processor chip, it has gone through three generations of iterations from procurement to self-research:
20 14 -20 16 Tesla is equipped with Mobileye? Automatic driving of EyeQ3? HW 1.0, including 1 front camera+1 millimeter wave radar+12 ultrasonic radar.
20 16 -20 19, driven by NVIDIA? PX？ 2? Autopilot of AI computing platform? HW2.0, and subsequent autopilot? HW2.5, including 8 cameras+1 millimeter wave radar+12 ultrasonic radar.
2065 438+09- Autopilot based on self-developed processor will be launched in recent years. What models were produced after the middle period of HW3.0 3.0,2019? S, model? X, model? 3 This FSD chip has been equipped with performance far beyond the second generation, and this generation has not made any changes on other sensors, but only upgraded the autopilot main chip (but it does not rule out the possibility of upgrading some sensors after the official announcement).
After the improvement of computing power, Tesla also launched a fully automatic driving preview version (FSD? Sneaking around? Preview), in this version, the system can accurately identify road/stop signs, trash cans, piles and barrels, lane lines and traffic lights. In other words, Tesla's next step is to open urban roads and fully assist driving.
Ok, the outline is clear, so how do the camera, millimeter wave radar and ultrasonic radar work together? We might as well get a general idea of the distribution without knowing the model and parameters of each sensor (we can't understand it either).
Tesla comes standard with eight cameras, which can be divided into three front-mounted three-eye cameras (telephoto, standard and fisheye), two B-pillar cameras, two fender rearview cameras and, of course, a rearview camera. Of course, if you have to count it all, models? There is also an in-car monitoring camera, but this monitoring camera has little contribution to the automatic driving function, mainly to monitor whether the driver's attention is concentrated during the automatic driving process.
● Front three-eye camera
● Front fender rear view camera
●B-pillar side camera
● Rear view camera
Through these eight cameras, Tesla has basically achieved 360 care without dead ends, making it possible to realize the automatic driving function.
After talking about the camera, this millimeter wave radar is also worthy of attention. From the early Bosch scheme to the mainland scheme of Autopilot2.5, the latter has improved in detection distance, weight and power consumption. However, at present, this set of hardware and algorithms is generally considered to have poor recognition ability for objects with strong lateral movement and reflection (refer to several truck accidents in which Tesla collided with lateral movement), so in the third generation of autonomous driving, this will be Tesla's key upgrade iteration direction.
● Millimeter Wave Radar on Tesla official website
Finally, there is this ultrasonic radar with as many as 12 pieces. Generally speaking, ultrasonic radar is the cheapest and the most common (such as the reversing radar installed in our car), but it is also the most vulnerable to environmental impact. It is difficult to develop in autonomous driving.
Generally speaking, we think that the detection distance of ultrasonic radar is very limited and it is too sensitive to the environment (especially temperature and humidity). Secondly, ultrasonic radar can't accurately confirm the specific position of obstacles, so the author thinks that these 12 ultrasonic radars mainly play an obstacle avoidance role when parking and overtaking.
After talking about the hardware and software configuration, we still have to fall into the actual experience. Through the deep integration of radar and camera (vision), Tesla has made a large-scale commercial autopilot assistance system with the fastest progress and the best experience so far. More importantly, Elon has announced: "At some point next year, Tesla owners can use the automatic driving function in their cars without human intervention."
This is not the distant future, and fully automatic driving is close at hand. Is Tesla the undisputed first person to drive automatically?
On financial strength and patent technology reserve: Waymo first
The reason why Tesla is said first is because it is the first time that users come into contact with fully automatic driving, but if you really want to be the first person in the field of automatic driving, you have to mention Waymo.
Waymo should be regarded as the most famous company in this field. The earliest Waymo was just a project department of Google. 20 16 officially became independent and became a company specializing in the field of autonomous driving. Moreover, Waymo itself is committed to the research and development of the autopilot algorithm layer and does not participate in building cars.
Waymo's method is to build an automatic driving system, which can be adapted by various car manufacturers, or these manufacturers simply provide their cars to Waymo to do it for them, so Waymo is more like providing a solution, no matter what car manufacturer you are and what kind of cars you have produced, you can actively adapt to this system.
Did you suddenly think of something when you saw this? Yes, this is the Android system in the mobile phone. Just as mobile phone manufacturers adapt to Android, car companies generally choose Waymo's solution. This solution is extremely large and complex, and only a giant like Google can complete it. Is it significantly different from the closed-loop mode of Tesla and Apple?
Since it's Google? Money is definitely no problem. After so many years of research and development, where has the technology gone?
2065438+February 2009, the California Motor Vehicle Authority (DMV) released an annual report, including 48 self-driving road test units with certificates in California, and Waymo ranked first, far ahead. This ranking mainly depends on the number of test vehicles used under the experimental conditions, the test mileage and the number of manual takeover during the test. Of course, the longer the mileage, the less frequent the manual takeover.
● Image source: Quantum bits
Waymo, which is committed to "one step in place", is much more radical than Tesla in hardware selection. The first is lidar. Waymo's lidar has its own transmitter, and it builds its own laser map through short distance, medium distance and long distance. The threshold of its camera is also higher, which not only has higher definition, but also enhances the recognition ability under strong light and dark light.
In addition, Waymo also chose electromagnetic wave radar, which is much more adaptable to the environment than ultrasonic waves, less affected by meteorological conditions and more stable. It's not over yet. Waymo even added many auxiliary sensors, including but not limited to voice recognition and GPS assistance. Is it much higher?
Nevertheless, after all, Waymo's test road conditions are all completed under experimental conditions, and its road complexity and samples are limited. The data accumulated by hundreds of test cars in one year, even with the newly opened Waymo? An unmanned taxi service is still a bit weak in the face of Tesla's huge delivery (of course, Tesla's autopilot function is gradually opening up, and there is no mouse problem in the previous acquisition).
Therefore, we believe that in a narrow sense, Waymo is far ahead in the number of patents in the field of autonomous driving, but in a broad sense, Tesla has more massive data and each has its own strengths.
And a hard blow from the main engine factory.
Except some big manufacturers will do it themselves (such as GM Super? Cruise), for a relatively small OEM, it is obviously not cost-effective to invest huge sums of money to independently develop autonomous driving.
For them, there is not much choice but to cooperate with upstream giant suppliers.
First, upstream suppliers complete hardware matching, and some car companies will develop software-based autopilot systems. However, due to the limitation of quantity and capital, they usually have to be integrated with the basic solutions of suppliers. In this mode, the development and iteration of autonomous driving ability is very difficult.
This is because this form of cooperation is generally dominated by upstream suppliers, resulting in neither side having much incentive to upgrade related technical capabilities, and suppliers will inevitably give priority to serving a larger number of partners; Therefore, in the field of autonomous driving, most car companies are standing still and lack innovation ability, which is nothing more than this.
In fact, from these three technical lines, we can probably understand why Tesla's autonomous driving technology has evolved so fast. The closed-loop R&D system similar to Apple also has dual drive capability of software and hardware, which brings a brand-new experience closest to fully automatic driving at present.
Of course, we are also looking forward to the day when Waymo has a big move, and the traditional host manufacturers who are still hesitating really have to work harder.
This article comes from car home, the author of the car manufacturer, and does not represent car home's position.