If last year's 1024 Xiaopeng Technology Day was exploring the possibility of future travel, and on yesterday's Technology Day, Xiaopeng seemed to have foreseeed the future and has completed the continuous evolution of intelligent robots and flying cars.

[ Pacific Auto Industry Channel] If last year's 1024 Xiaopeng Technology Day was exploring the possibility of future travel, and on yesterday's Technology Day, Xiaopeng seemed to have foreseeed the future and has completed the continuous evolution of intelligent robots and flying cars. With the mass production of the 800V high-voltage SiC platform model, the super energy supplement system only occupies one PPT, and spends more time sharing Xiaopeng’s technical thoughts on “urban NGP” and “XNGP”, as well as the display of intelligent interaction capabilities.

XNGP - the ultimate technical form of assisted driving

Starting from September 17, the urban NGP intelligent navigation assisted driving began to be piloted in Guangzhou, and was successively pushed to P5 users. The Xiaopeng Motors , which insists on the full stack self-developed , has become the first automobile brand to allow advanced intelligent assisted driving to be mass-produced and implemented in the urban market scenario.

Dr. Wu Xinzhou, Vice President of Automobile Automobile, once again emphasized the significance of the urban market scenario to the development of intelligent driving. The implementation of the urban market scenario last year marked the "start of the second half of intelligent assisted driving". The implementation of the urban market scenario this year is a decisive step towards the second half. Not only for Xiaopeng, but Xiaomi also regards autonomous driving as the first breakthrough direction.

Compared with high-speed NGP, the number of codes in urban NGP has increased to 6 times, the number of perceived models has increased to 4 times, and the number of prediction/planning/control related codes has increased to 88 times. According to data released by Xiaopeng, the current urban NGP weekly user penetration rate has reached 84%, the mileage penetration rate has reached 63%, and the traffic efficiency can be close to 90% of human drivers. Passive takeovers per 100 kilometers are only 0.6 times. The implementation of

Urban NGP marks the completion of the branch of full-scene intelligent assisted driving (XNGP) and the XNGP equipped with Xiaopeng G9 (query the lowest price | check the matching) is also the ultimate technical form of assisted driving before unmanned driving.

XNGP not only does not require high-precision maps, but can be used in all cities and sections of roads across the country. It can also cover daily commuting and seamlessly connect various scenarios such as highways, cities, underground parking lots, etc., truly forming a closed loop, and at the same time strive to achieve zero takeover in 2023.

XNGP's capabilities will rely on powerful hardware, software that upgrades capabilities, and fully closed-loop, self-grown AI and data systems. Among them, Xiaopeng introduced the new XNet deep vision neural network technology. Compared with Xiaopeng Motors' first generation visual perception architecture, it used neural network to replace complex handwritten logic, realizing end-to-end data-driven algorithm iteration.

It can fuse the data collected by multiple cameras in multiple frame timing pre-fusion, output 4D information of dynamic targets from the perspective of BEV (such as the size, distance, location and speed, behavior prediction, etc.), as well as 3D information of static targets (such as the location of lane lines and road edges).

In addition, Xiaopeng Motors has built China's largest autonomous driving intelligent computing center in Ulanqab . Based on Alibaba Cloud intelligent computing platform, the computing power can reach 600PFLOPS ( floating point calculation billion times per second). With the support of "Fuyao", the training efficiency of autonomous driving models has been increased by 602 times. Compared with the stand-alone training of a server with 2400TOPS computing power, 80-machine parallel training can shorten the training time from 276 days to 11 hours.

The basic capabilities of data closed loop consist of four core abilities: collection, annotation, training, and deployment. Efficiency is one of the important assessments of the AI ​​system. Through directional collection, Xiaopeng Motors can collect a large amount of training data through massive vehicles, and quickly correct it with the help of automatic labeling system and Fuyao.

XNGP, XPILOT latest plan

XIPOT also brought the latest XNGP and XPILOT release plans, which are more specific than last year's plan.

1. From 2022 to the first half of 2023, the Xiaopeng G9 Pro version will open traffic light recognition and lane-level navigation functions based on the functions of high-speed NGP, memory parking, LCC, smart parking and other functions as soon as it is delivered.

2. This function is in Xiaopeng P5P version and E version, Xiaopeng P7E version and Zhizun version will also be opened. Xiaopeng P5 P version of the city NGP will add new cities to Shenzhen and Shanghai, and the performance of the city NGP and LCC enhanced versions will also be further improved.

3. Before the first half of 2023, in addition to the G9 Max version, which will have high-speed NGP, memory parking, LCC, smart parking and other functions, the ability to identify traffic lights and pass through intersections nationwide (all cities without pictures) will be open to the ability to directly pass through intersections. The urban NGP covers three cities, Guangzhou, Shenzhen and Shanghai.

4. In the second half of 2023, the high-speed NGP, VPA, and LCC of the Xiaopeng G9 Pro version will usher in performance improvements. The high-speed NGP functions of all the above versions will also usher in strategy optimization in speed limit adjustment and hand-off detection during this cycle.

5. In the second half of 2023, lane change, overtaking, and left-right rotation capabilities will be opened in most cities without maps (the available range of high-level capabilities will be opened in stages based on the number of users and the test progress).

6, complete the full scenario opening in 2024, realizing intelligent navigation and assisted driving capabilities from parking spaces to parking spaces.

At the same time, Xiaopeng will realize the closed loop of data and capabilities through dual-line parallelism between autonomous driving and high-level assisted driving. Judging from the current plan, 2022 will be single-scene assisted driving; 2023~2025 will be full-scene assisted driving, realizing assisted driving from parking space to parking space; after 2025, it will be fully autonomous driving. Judging from the FLAG established by Xiaopeng, can it be mass-produced?

Full-scene Voice 2.0

Two years ago, Xiaopeng launched the industry's first "full-scene Voice". Two years later, Xiaopeng brought the industry's first full-vehicle and full-time voice interaction function - Full-scene Voice 2.0.

For the first time, it applies MIMO multi-sound zone technology to the car voice system, realizing that the small P is on standby throughout the entire journey. Multiple occupants in the car do not need to wake up, and multiple people are talking to each other at the same time.

Xiaopeng's self-developed voice interaction architecture can realize parallel processing of multi-channel voice streams in the whole vehicle, and supports the processing of voice interaction requests interwoven with various functions such as concurrency in four sound zones, integrated end-cloud, real-time stream recognition and understanding, parallel commands, etc.

can also understand the multi-channel context in the whole vehicle. It not only maintains single-person dialogue and context management, but also compatible with multi-person dialogue and context understanding, and resolves reply conflicts between multiple voice channels and four-channel voice interactions.

G9 also efficiently uses local computing power to realize local dialogue, and can also stably control more than 600 vehicle functions under weak and no network conditions.

"Little White Dragon" continues to evolve into humans

Xiaopeng Pengxing's intelligent robot "Little White Dragon" has also ushered in a new generation of prototypes, with a very sci-fi appearance and a futuristic feel. The head integrates AR projection, which is an exploration of robot interaction forms, and seems to be more anthropomorphic than " Iron Egg ".

The soles of the foot use EVA supercritical foaming and wear-resistant rubber , providing excellent cushioning, silent and wear resistance; new elastic fabrics and liquid silicone materials are used in the neck and other areas, solving the design problem of multi-degree of freedom movement.

In addition to the upgraded shape, Xiaobailong's architecture has also been upgraded. The bones on the body and legs are designed with a lightweight design, and the computing unit inside the robot is the same as the smart car's computing power; it is equipped with a 6-axis cooperative robot arm, which cooperates with a four-legged chassis to perform more functions in the home environment.

As He Xiaopeng said before, the machine is autonomous driving as soon as it comes, and there is no stage of "human-machine driving together". The environment to be identified is complex and the goals are diverse. At the same time, we must also perceive a three-dimensional environment, such as going up and down stairs and running on all terrain. The interaction with people and the environment is more complex and has a strong dependence on vision and spatial cognition.

More importantly, there is the perception of smell and touch, which enables machine horse to recognize various odors like humans or animals, and can also perceive various touch interaction forms between users and machine horses, such as soothing the head and contacting the "skin", which can provide emotional expression.

Intelligent robots and smart cars are more like complex hardware and software integrated products. They achieve each other and have different focuses, but technical coordination can be achieved, but the discussion of the best form is still in progress.

Flying Car - Let the car fly

The shape of Xiaopeng Flying Car has also been adjusted, and it has an arm folding storage system, which can switch between land travel and flight modes.In the folding and storage state of the machine, its size is comparable to that of a conventional car and can drive freely on open roads.

Compared with last year's concept model, the latest design is optimized from the horizontal dual rotor configuration to a distributed multi-rotor configuration. The current solution is shorter arms and blades, the total length and width of the flight state are smaller, the size limit on the take-off and landing site, and the flight passability is better.

At the same time, the paddle folding and periodic variation paddle hub are cancelled, making the folding and storage method simpler. The overall system design complexity is reduced, and safety and reliability are improved while achieving lower noise levels.

In addition, it has the fault-tolerant control function of flight control and dual backup of power. Even if a single point of failure occurs, it can still maintain controllable flight and safe landing. It has successfully conducted multiple single-motor failure tests on the early exploration products.

Xiaopeng Huitian's 2-ton test prototype has completed its first flight, which is also the world's first real electric hanging takeoff car. Judging from the current form, it is not very similar to a brand new flying car, but allows the car to fly.

Flying cars will also innovatively adopt Xiaopeng Huitian's self-developed driving control system, which couples various functions of aircraft and car control to meet the driving needs of both flight and land travel.

In the future, flying cars will continue to upgrade their range, because "flying resistance" is much greater than the car's driving resistance. Xiaopeng Huitian needs to develop energy solutions with higher energy density and higher power density, and will also achieve technical coordination with Xiaopeng Motors.

At the same time, autonomous driving is also a key step, and it must be intelligent and simple enough to learn and use, including automatic flight cruise. At the same time, flying cars have the same perception hardware when flying or land. Using the perception hardware of the flight state in the land state will be a complete dimensionality reduction blow for pure cars on land.

written at the end

From the original intelligent technology iconic 1024 Technology Day, Xiaopeng has been constantly using mass production as the purpose, including the city NGP that has been implemented, creating unique technological competitive advantages. On the 1024 Technology Day last year, if you think that the flying cars mass-produced in 2024 are still imaginative and out of reach, but on this year's Technology Day, I gradually chose to believe that because we can predict the flying car , and we also look forward to the three-dimensional transportation layout that can arrive earlier!

(text: Pacific Auto Yayong)