The fourth "1024 Xiaopeng Auto Science and Technology Day" came as scheduled, with the theme of "from foresight to more than meeting", sharing the exploration process and mass production promotion plan of Xiaopeng Auto in the future travel field in the four dimensions of intelligent driving, intelligent interaction, intelligent robots and flying cars.

"1024 Xiaopeng Automobile Science and Technology Day", see what black technologies will be in the future? 1024 XPENG TECH DAY in 2022

The implementation of urban scenes is the key to the development plan of intelligent driving assistance system

"1024 Xiaopeng Automobile Science and Technology Day", see what black technologies will be in the future?

The Future Development of Xiaopeng Intelligent Assisted Driving

The landing of the urban scene is just the beginning of the second half of intelligent auxiliary driving. The XNGP intelligent auxiliary driving system carried by Xiaopeng G9 has the capability of intelligent auxiliary driving in all scenes, and is also the ultimate technical form of auxiliary driving before unmanned driving. It can be used in all cities and road sections of the country without high precision maps. It can also cover daily commuting and seamlessly connect various scenes such as high-speed, urban and underground parking lots. The experience of XNGP will also exceed that of most drivers in terms of safety, road game ability and traffic efficiency. In order to achieve this capability, Xiaopeng Automobile has significantly upgraded its hardware, adopted 508 TOP computing power+dual laser radar+800W pixel high-definition camera, and applied the software architecture and full closed loop, self growing AI and data systems that have been improved again.

"1024 Xiaopeng Automobile Science and Technology Day", see what black technologies will be in the future? XNGP intelligent auxiliary driving system

Among them, XNet depth visual neural network will fuse the data collected by multiple cameras before time sequence, and output the 4D information of dynamic objects (such as the size, distance, position and speed of vehicles, two wheeled vehicles, etc., behavior prediction, etc.) under the BEV perspective, as well as the 3D information of static objects (such as the position of lane lines and road edges). Compared with the first generation visual perception architecture of Xiaopeng Automobile, XNet uses neural networks to replace complicated handwriting logic and realizes end-to-end data driven algorithm iteration. Of course, behind the excellent technology is the huge data collection, annotation, training and spacing deployment. Taking 500000 pieces of training data Clip and 1 billion objects as an example, it takes 2000 person years to manually label them, and 276 days to conduct stand-alone training with a 2400TFLOPS computing server. At the same time, if it is deployed to the vehicle end without optimization, this single network alone can occupy 122% of the computing power of a single Orin-X. Therefore, Xiaopeng Automobile has built a fully automatic labeling system, which is 45000 times more efficient than manual labeling. That is, XNet requires 2000 person years of labeling. Now it only takes 16.7 days to complete the labeling. In addition, Xiaopeng Automobile has built China's largest automatic driving intelligent computing center "Fuyao" in Ulanqab. Based on Alibaba Cloud intelligent computing platform, its computing power can reach 600PFLOPS (6 billion floating point operations per second). With the help of "Fuyao", the training efficiency of automatic driving model has been improved by 602 times. Compared with the single machine training of 2400 TOPS servers, the 80 machine parallel training can shorten the training time from 276 days to 11 hours. At the same time, the ultimate engineering capability accumulated from the research and development of urban NGPs allows 122% of Orin-X computing power required by dynamic XNet to be only 9% after optimization. The ultimate engineering deployment capability is the key to the launch of XNet, and also the key to the continuous iteration in the future.

"1024 Xiaopeng Automobile Science and Technology Day", see what black technologies will be in the future?

XNet Deep Vision Neural Network

In addition, the basic capability of data closed-loop consists of four core capabilities: collection, labeling, training and deployment. Efficiency is one of the important KPIs of the AI system. Therefore, Xiaopeng Automobile can collect a large amount of training data through a large number of vehicles through directional collection, and quickly make corrections with the help of the automatic marking system and Fuyao. Even for some rare extreme scenes, the simulation system can generate a large number of data of the same type for rapid iterative training. Through the combination of directional acquisition and simulation, more than 1000 Corner cases (extreme scenarios) are solved in one year, and the high-speed NGP accident rate is reduced by 95%. At the same time, every line of code change needs to go through a simulation mileage of 50 million kilometers, 5000+core simulation scenarios, and 17000+special simulation scenarios to ensure the reliability and positive evolution of function iteration.

Wu Xinzhou, Vice President of Xiaopeng Auto Driving, delivered a speech

With the high efficiency brought by the software and hardware system, the XNGP system will, before the first half of 2023, be able to recognize traffic lights and directly pass through the road crossings nationwide (all cities without maps), except for the high-speed NGP, memory parking, LCC, intelligent parking and other functions that will be available when the G9 Max model is delivered. The urban NGP covers Guangzhou, Shenzhen and Shanghai. In the second half of 2023, lane changing, overtaking, left and right turning capabilities will be opened in some cities without maps (the available range of some high-level capabilities will be opened in stages according to the number of users and test progress). In 2024, the whole scene will be opened, and the intelligent navigation assisted driving capability from parking space to parking space will be realized. In addition, for users of XPILOT version, Xiaopeng's intelligent assistant driving ability will also be continuously updated. From 2022 to the first half of 2023, Xiaopeng G9 Pro will open the traffic light recognition and lane level navigation functions on the basis of the instant delivery of high-speed NGP, memory parking, LCC, intelligent parking and other functions. This function will also be opened in Xiaopeng P5 P and E, Xiaopeng P7 E and Zhizun. The urban NGP of Xiaopeng P5 P will be newly opened in Shenzhen and Shanghai, and the performance of urban NGP and LCC enhanced version will also be further improved. In the second half of 2023, the high-speed NGP, VPA and LCC of Xiaopeng G9 Pro will welcome the performance improvement, and the high-speed NGP functions of all the above models will also welcome the strategy optimization in speed limit adjustment and release detection in this cycle. (Note: The urban NGP urban opening progress is subject to the OTA plan, and some functional upgrades need to have corresponding software and services)

XNGP release plan

XPILOT Release Plan

The arrival of Xiaopeng's new generation of intelligent robots

Since its debut in 2021, Xiaopeng Pengxing's intelligent robot has continued to explore and ushered in a new generation of prototypes. Compared with the previous generation, the current model is more sci-fi and has a future style: the head is integrated with AR projection, which is an exploration of robot interaction forms; The same "EVA supercritical foam" and wear-resistant rubber as high-end running shoes are used on the sole to provide excellent cushioning, mute and wear-resistant performance; New elastic fabrics and liquid silica gel materials are used in the neck and other areas to solve the design problem of multi degree of freedom movement, and they are impact resistant and scratch resistant.

Xiaopeng Pengxing New Generation Intelligent Robot

In addition to the model upgrade, the "Little White Dragon" also welcomes the architecture upgrade. The "bones" of the body and legs refer to the lightweight design ideas in the automotive field, which are stronger and lighter than the early generation; The new generation of AI computing units and intelligent cars have the same generation of computing power, greatly improving the intelligent potential of robots; At the same time, it is also equipped with a high fidelity sound system to bring exquisite voice and audio-visual entertainment enjoyment; Of course, the most significant point is that the upgraded "Little White Dragon" is equipped with a 6-axis cooperative mechanical arm, which can cooperate with the quadruped chassis to perform multiple functions in the home environment, greatly improving the potential of robot interaction with the physical environment.

Under the Xiaopeng ecosystem, during the research and development of intelligent robots, the power system, automatic driving, human-computer interaction, research and development system and other technologies have been effectively empowered from the field of intelligent vehicle research and development, including power system, automatic driving, human-computer interaction and AI capabilities. AI capability will become the "engine" driving the development of robot intelligence and automobile intelligence technology, data will become the "energy" driving the development of technology, and automobile intelligence and robot intelligence will "achieve each other". At present, the discussion on the best form of robot is still in progress, and Pengxing's products may have a new form in the future.

"Road air integration" has become possible, and the world's first electric suspension flying car

On 1024 Science and Technology Day in 2021, He Xiaopeng brought the concept model of flying car. Over the past year, on the basis of the original, Xiaopeng Huitian has also been promoting mass production. In this year's 1024, Xiaopeng Huitian released the latest configuration of the flying car, which is equipped with an arm folding storage system and can switch between land and flight modes. In the folding and storage state of the boom, its size is equivalent to that of conventional cars, and it can run freely on open roads; It can switch between land mode and flight mode. Take off and land vertically under the conditions allowed by laws and regulations and the environment, and fly across congestion, obstacles, rivers, etc. to meet the needs of people for short distance low altitude travel.

Land/flight status of Xiaopeng Huitian flying car

Compared with the concept model last year, the latest design has been optimized from the tandem twin rotor configuration to the distributed multi rotor configuration. In the current scheme, the boom and blade are shorter, the total length and width of the flight state are smaller, the size of the takeoff and landing site is less limited, the flight accessibility is better, and the use is more convenient. At the same time, blade folding and periodic variable pitch hub are canceled, which makes folding and storage easier. The overall system design complexity is reduced, and the safety and reliability are improved while achieving lower noise level. In addition, it has the fault-tolerant control function of flight control and dual backup of power. Even in the event of a single point failure, it can still maintain controllable flight and safe landing. At present, Xiaopeng Huitian has successfully conducted multiple single motor failure tests on early exploration products.

The flying car will also innovatively adopt the flying car driving control system developed by Xiaopeng Huitian. It couples various functions of aircraft and vehicle control to meet the requirements of flight and land driving. In the land state, the driving mode is basically the same as that of existing vehicles; In the flight state, through the cooperation of the steering wheel and the right gear lever, it is easy to realize the forward, backward, turning, ascending, hovering, descending, etc. of the flying car in the air.