On August 27th, at Xiaopeng’s 10-year love night and Xiaopeng MONA MO3 listing conference, he Xiaopeng announced that Xiaopeng’s self-research Turing chip had been successfully released on August 23rd.
He Xiaopeng confirmed that the flying car, which integrates ten years of research and development, will officially go on sale in the fourth quarter of this year.
Xiaopeng AI robot will be unveiled on October 24th.
, according to reports, Xiaopeng Turing chip is the world’s first AI chip used in AI cars, robots and flying cars, customized for AI models.
For L4 autopilot, using 40 core processors, integrated 2xNPU self-developed neural network to deal with the brain, for L4 autopilot design, the computing power is three times that of the existing chips.
The effective visual perception information of the end-to-end large model is increased by 8 times, the spatio-temporal sequence is improved, and the environmental understanding and prediction ability is improved, which greatly improves the visual perception ability.
In addition, the chip is also equipped with an independent safety island, which can realize real-time blind spot monitoring of the whole vehicle.
Built-in 2 independent image ISP, up to run large model parameters 30B.
Dark night, rainy day, backlight are all clear.
, 
, in addition, Xiaopeng released a new pure visual smart driving program– AI Hawkeye.
It is said that the AI Eagle Vision solution is clearer than the human eye, using the global premiere Lofic architecture, camera accuracy, visual range upgrade, color resolution is more enhanced, in low light, backlight, large light difference and other environments can see clearly, in the dark, rainy and foggy days, tunnels and so on can be clearly identified.
, 
, AI Eagle Vision solution calculates 20%, can be used to increase the capacity of large models and the number of parameters, visual information without information conversion, directly in the neural network as input.
The real-time perceptual distance is 1.
25 times higher than that of the previous generation, the recognition speed is increased by 40%, the delay is reduced by 100ms, the perceptual regulation is more sensitive, the response is faster, the passable space is output directly, the path planning information is more direct, and the response is faster.
What is worth being integrated is that the Xiaopeng P7 + will be the first model in the world to be equipped with the AI vision Eagle Vision solution, while the previous models equipped with the lidar scheme will be synchronously updated with OTA to maintain a consistent experience.
, 
, Xiaopeng intelligent driving will be divided into four parts, in which the first step has been landed, and the light map is still open all over the country.
the second step: 2024 Q4 to achieve light radar, light map, door-to-door intelligent driving experience, large model parameters x2.
third step: H2 to achieve L3-like intelligent driving assistance in 2025, 100 km takeover < 1, cloud large model parameters x5 Step 4: in some low-speed scenarios in 2026, people will not drive in the driver's seat, solving 99% of the extreme problems.
The end-to-end large model increases the utilization rate of visual perception information by 8 times, enhances the end-to-end spatio-temporal understanding ability, and improves the environmental understanding ability through longer time series information and more spatial information, which can not be seen by the information “brain supplement”.
Higher precision visual information, not only a higher upper limit, but also a safe bottom, the self-driving experience will improve quickly.
, the first model of Xiaopeng MONA series, M03, was officially launched, with a total of three configured models with a price range of 11.
98-155800.
Xiaopeng MONA M03 will be officially delivered to users in September, and the Max version will be delivered after the Spring Festival in 2025.
During the listing period, the whole department gave away Xiaopeng Mall points, 4000 yuan PSS customized 18 speakers free upgrade and delivery gift package.
, 
, return to the first electric network home page >.
