10-24, a technology day for Xiaopeng is here.
This time, there are three main themes, which are still focused on intelligence:
- City NGP
- Intelligent robot
- Flying cars
Next, let's get to the point.
City NGP
City NGOP delivery, enter the second half of intelligent assisted driving
Although the urban market scenario is difficult to do, Xiaopeng must firmly plunge into:
Because in Xiaopeng's view, this scenario is "high frequency + urgent need", Wu Xinzhou said: Users are willing to pay for this scenario.
Urban NGP development code volume is 6 times that of high-speed scenarios, the number of perceived models has increased by 4 times, and the difficulty of R&D has increased by 100 times.
Specific improvements: horizontal control, driving game between mattes and mattes, and handling of uncertain scenarios.
In addition, you are more skilled in map applications:
This is very interesting. In the absence of high-precision coverage/high-precision map information is different from the actual driving scenario. NGP will be downgraded to LCC and used visual identification and processing. After passing this section, you will continue to drive back to NGP, and this transition will become more indifferent in the future.
This is similar to the previous high-speed NGP logic, but the experience remains to be tested.
also proves from the side that Xiaopeng City NGP is similar to high-speed NGP and is highly dependent on high-precision maps. What can be previewed by is that in the scene mentioned above, the urban NGP experience will be somewhat separated.
9/17, Xiaopeng City NGP pushed it to internal employees, and conducted public beta to hundreds of users in states on October 8, and on October 22, it was fully opened in states.
During this period, the achievements made by urban NGP:
Urban NGP assisted driving penetration rate 84%. What is this concept? 3-4 times higher than previous LCC.
urban NGP route penetration rate reaches 63%, that is, every 10 public cities can start the urban NGP route, and users have an average of 6 publics using this function.
In addition, in 2023, Xiaopeng also drew a flag: can really achieve 0 takeover at high speed. After the city market scenario is implemented, the next step is to implement all-scene assisted driving. Let’s look at the timetable first:
2023-2025, realize all-scene assisted driving;
2025-In the future, we will move towards full-scene autonomous driving and unmanned driving.
All-scene assisted driving will be named XNGP. X, represents infinite possibilities. XNGP=High-speed NGP+Stop NGP+Urban NGP.
2025 will have a new software and hardware architecture. At that time, Xiaopeng will move towards comprehensive autonomous driving and unmanned driving.
XNet: Speed perception and intention prediction of pure vision
XNet behind XNet's excellent technology is huge data collection, annotation, training and arduous deployment
XNet Currently, there is 1,000 people in the XNet tagging team.
In addition, Xiaopeng also has its own fully automatic labeling system, with efficiency increased by nearly 45,000 times. The original 2,000-year labeling volume can now be completed in 16.7 days:
Xipeng’s self-driving training center:
Xipeng Transformer Deploy
XNGP's fully closed-loop, self-contained AI and data system:
XNGP's release plan (currently mainly targeting G9 type):
In the first half of 2023, about one week, Xiaopeng City NGP will be implemented in G9 and in Shanghai and Shenzhen Prefecture.
In addition, the ability to identify traffic lights and perform cross-traffic crossings is open nationwide (cities without pictures).
In the second half of 2023, most cities without maps have the ability to approach the city GDP.
2024 will achieve full-scene connectivity.
XPILOT Release Plan:
For G9 Pro Model: In the first half of 2023, we will open traffic light recognition and channel-level navigation.
P5 lidar version: urban NGP will be expanded to at least 3 cities in the first half of next year. Optimization of high-speed NGP strategies in the second half of the year.
One more thing: Xiaopeng Robotaxi is coming. A domestic zero-modified mass production of autonomous driving closed road test has obtained the "intelligent connected vehicle road test license.
Xipeng's autonomous driving idea: assisted driving and unmanned driving parallel to realize the closed loop of data and capabilities.
However, as for what she said last year, "In the second half of 2022, it will become a Chinese entire
enterprise that Chinese people explore autonomous driving travel operations."
may continue to be postponed.
Xipeng full-scene voice 2.0:
wake-up
does not affect each other:
Reaction speed is very fast:
can be used without network:
Full stack self-developed voice architecture:
Intelligent robot
Emmmm, it is more suitable to call a robot dog.
Here, Xiaopeng has done a lot of explorations: AR projection, advanced audio, natural dialogue...
As for other improvements, including improving the silence of the robot dog's movement and obstacle avoidance ability, and then adding an robot arm ...
greatly accelerates the technological progress with system capabilities: putting the ability of the robot (this idea is increasingly like Tesla )
people feel that the technical improvements in the hard meme this year are not much, but more concentrated on the application of the functional level. This little Dong is not able to arouse my desire for the time being.
在彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩彩�
Improvements made in the past year: Using a distributed multi-rotor configuration, it takes half a basketball court to expand
1
Xipeng Huitian's self-developed umbrella motor driving control system
Test video screenshot:
College Automobile Challenge:
1
101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010100101010101010010101010100101010100101010100101010010
is written at the end
or above, which is the entire content of this press conference. Looking forward to next year's 1024.
If you think the content is good, please like, follow and forward it. This will be of great help to my creation.