With Tesla, Weilai, Xiaopeng, Ideal and Great Wall and other OEMs entering the high-speed pilot assistance, more and more consumers have also experienced the convenience of intelligent driving. Later, with the arms race of various car companies in the field of intelligent driving, "1, 2, 3, 4"... The lidar ride seems to indicate that the arrival of the L4+ is speeding up.
From "assisted driving" to "autonomous driving", which step should we take first? And what route should we take? Let's listen to the answer sheet handed over by Michiyuki.
Come on, show
The autonomous driving company, which was independently formed by the predecessor "Great Wall Motor Intelligent Driving Development Department", is currently doing "passenger car assisted driving" and "low-speed unmanned vehicle ecosystem". At the 2022 HAOMO AI DAY, The First Wave of Report Cards was shown:
In May 2021, HWA high-speed driver assistance system was mass-produced;
In December 2021, the NOH Intelligent Pilot Assisted Driving System was mass-produced, and the first model was the Weipai Mocha NOH Smart Pilot Edition, and I had also experienced engineering prototypes before:
In December 2021, the data intelligence system MANA was released, and as of now, MANA has completed nearly 200,000 hours of learning time, and the virtual driving age is equivalent to 20,000 years of human drivers;
By the end of 2021, the intelligent driving system has been installed on the 5 models of Weipai Mocha, Weipai Latte, Weipai Machiduo, Haval Divine Beast, tank 300 City Edition, with a load of more than 10,000 units;
In March 2022, the 6th model tank 500 equipped with the millime-terminal driving system was officially launched;
As of April 2022, the user mileage of the millima intelligent driving system has exceeded 7 million kilometers.
The noh of the city was released for the first time
At HAOMO AI DAY, Miller released the "NoH of Milliped City" equipped with HPilot3.0. Millima City NOH is an intelligent driving system that "assists" drivers to complete driving based on user-set navigation routes in urban road scenarios, and at present, Millima City NOH has completed all functional development.
(Millime City NOH Test Vehicle)
Weipai Mocha will become the first model equipped with noH in the city of Milli, and the specific landing model information is as follows:
Perception hardware: equipped with 2 lidar, 5 mm radar, 12 high-definition cameras and 12 ultrasonic radars;
Computing platform: Equipped with the autonomous driving computing platform "Little Magic Box 3.0" jointly launched by Millai Zhixing and Qualcomm, the chip AI computing power is 360TOPS, the cache is 144M, and the CPU computing power reaches 200K+ DMIPS;
Scene coverage: According to the driving route provided by the navigation, the noh of the city can realize the main functions of automatic lane change overtaking, traffic light recognition and control, complex intersection traffic, unprotected left and right turning in the urban environment, and can also cope with complex urban traffic scenarios such as vehicle close entry, vehicle blockage, intersection, roundabout, tunnel, overpass and so on.
According to the information given by Miller, under the chaotic state of Level 4 (that is, the slow-moving scene and insufficient traffic space), the intersection pass rate of noH in Millima City is 70%, and the success rate of lane change is 90%.
Two days ago, I also went to experience the test car of noH in the city of Milli (the test car is equipped with 3 lidars):
According to the planning of Milli- end, in 2024, there will be more than 100 cities where the "Millipede City NOH" function will be landed, and the number of passenger cars equipped with the system will exceed 1 million units, and at the same time, high-speed, urban expressway, urban open road, parking lot and other scenarios will be opened.
MANA is upgraded again
Since the release of the data intelligence system MANA in December last year, in four months, MANA has ushered in a number of upgrades in terms of reducing costs and improving iteration speed.
Perception intelligence: The "double-stream" perception model is used to decompose the traffic light detection and the road binding problem into two channels, so that the daily passenger car test at the end of the millimeter realizes the traffic light recognition under heavy perception. The self-developed BEV Transfomer is used for lane line recognition, which automates labeling through two stages of "target coarse positioning" and "fine attribute estimation", and realizes multi-sensor fusion lane line recognition on urban roads.
Cognitive intelligence: MANA replaces the traditional handwriting rules and parameters with machine learning models, solves the problem of bloated code and easy to collapse and fail in the face of complex scenarios, makes scene decisions more generalizable and applicable, and improves interpretability and generalization capabilities.
Cost and Speed: The PAI-EFLOPS team and alibaba cloud PAI-EFLOPS team cooperated to implement distributed training of Swin Transformer models based on 128 card A100 clusters. Together with the Alibaba team, we have jointly innovated technologies including hybrid precision optimization, operator optimization, compilation optimization, etc., and through the optimization of large model training, the current model training cost has been reduced by 60%, the acceleration ratio has exceeded 96%, and the throughput has exceeded 40,000 samples per second. In addition, the automation rate of annotation AI has reached 80%, which improves the annotation efficiency and reduces the cost of annotation costs.
At the same time, in terms of data security, privacy protection and data security guarantees have been added to the existing processing network to protect data security.
Start a large-scale war on unmanned vehicles
The second generation of unmanned delivery vehicles released at AI DAY, Little Devil Camel 2.0, is the industry's first 100,000-level terminal logistics automatic delivery vehicle for the commercial market, and is expected to be released on the market in May 2022.
Little Magic Camel 2.0 is based on the general-purpose wire control chassis "Little Magic Disk" developed by Millima:
Little Magic Camel 2.0 is equipped with the Little Magic Box 3.0 computing platform, which has L4 automatic driving capabilities and covers complex traffic scenarios such as mixed traffic and congestion;
The design speed is 40km/h, and the working temperature zone is -20°C~45°C, which supports the medium and low speed of urban open roads and the high and low temperature all seasons.
Equipped with a 3.6kwh/6kwh battery pack, it can provide a cruising range of 60km/100km;
Equipped with a 600L cargo case, it also supports functions such as fast power change, intelligent voice and multi-mode interaction.
At present, the terminal logistics automatic distribution vehicle factory in Baoding has been fully upgraded, covering an area of 10,000 square meters, which can achieve the production capacity target of 10,000 terminal logistics automatic distribution vehicles per year.
Obviously, in the field of automated delivery vehicles for terminal logistics. I want to start a "unmanned vehicle scale war" with the advantages of "low price, strong performance, full scene, and high production capacity".
At last
In addition to the Wei brand, the follow-up Haval and tank brand models will also be gradually equipped with the noH system. By the end of this year, the millimeter intelligent driving system is expected to land 34 models. In 2023, the HSD Self-Driving will be officially launched.
This year, smart driving officially opened the second half of the showdown.
In the field of commercial vehicles, the first half of Robotaxi has long ended, and "large-scale commercial operation" has become the key word in the development of the industry. In the field of passenger cars, with the increase in penetration rate, the battle situation of the pilot auxiliary driving system began to shift from high-speed and expressway to urban road scenes.
Although it is a "rising star" in the field of intelligent driving, it seems that it is catching up at a very fast speed. From the software and hardware of the intelligent driving system to the data intelligence system, from the passenger car to the unmanned vehicle to the future Robotaxi, the current millimeter has completed their layout in the field of intelligent driving.