Xpeng Sensing Strategy for NOA & L2+ Systems
At least 10 chinese automakers and suppliers have launched driver-assistance systems over the past two years that can navigate city streets and make turns at intersections. EV makers Xpeng, Li Auto, BYD, Leapmotor, Xiaomi, Nio, and Huawei are among the carmakers with L2 or L2+ driver systems. Any new model priced at more than the equivalent of USD $30,000 in China now needs advanced driver-assistance features to compete, said Maxwell Zhou, co-founder of DeepRoute, a China-based startup selling software for advanced driver-assistance systems. Carmakers in China have two options: develop the systems themselves, or form alliances with suppliers.
Before Huawei started putting lidars on cars, Xpeng was the original autonomous driving leader in China. Their P5 was the first production car in the world to be fitted with lidar and boast city autonomous driving, but only in one city.
Xpeng is one of the Chinese startups in the field of NEVs (New Energy Vehicles), initially competing with Nio and Li Auto. Below you’ll find the ranking of Chinese EV makers for November 2024. For reference Tesla’s Gigafactory Shanghai delivered 78,856 vehicles in November 2024.
XPeng Competitive Position
The new Xpeng Mona M03 surpassed 10,000 units in deliveries for two consecutive months since its launch.
Xpeng has reported a third-quarter revenue of C¥ 10.1bn ($1.44bn), beating analysts’ estimates of C¥ 9.9bn in a Bloomberg survey, with a gross margin hitting a record high. Xpeng’s net loss was C¥ 1.8bn ($256m), compared to C¥ 3.9bn for the same period in 2023 and C¥ 1.3bn for the second quarter of 2024. Xpeng’s R&D expenses for the third quarter were a record C¥ 1.6bn, an increase of 25 percent compared to C¥ 1.3bn for the same period in 2023.
For the same quarter, Nio reported vehicle sales of C¥ 16.7bn (US$2,379.4m) and a net loss of C¥ 5.0bn (US$721.0m). Research and development expenses in the third quarter were C¥ 3.3bn (US$472.9m).
After the recent launch of the Mona M03, Xpeng now has quite a large family of car models including sedans, SUVs, and MPVs.
XPeng Mona M03 – 2024 – Compact sedan
XPeng P5 – 2021 – Compact sedan
XPeng P7 – 2020 – Mid-size sedan
XPeng P7+ – 2024 – Full size sedan
XPeng G9 – 2022 – Full-size SUV
XPeng G6 – 2023 – Mid-size SUV
XPeng X9 – 2024 – Full-size MPV
XPENG Experience in Intelligent Driving
Xpeng is one of the NEV brands famous for its intelligent driving features, and here is a look at that history.
In April 2018, Xpeng unveiled their G3 all-scenario automatic parking technology, Xpilot 2.0, with 20 sensors. This vehicle introduced all-scenario intelligent parking through the fusion of visual sensors and ultrasonic sensors. It is the G3’s function of automatic parking that branded Xpeng as a leader in intelligent technology, making it one of the closest competitors to Tesla in China.
In June 2018, Xpeng and Desay officially signed a strategic cooperation agreement in Guangzhou. Their collaboration on the autonomous driving system aimed to deliver three core automated driving functions: low-speed valet parking, mid-speed traffic jam assisted driving, and high-speed autonomous driving.
In 2020, Xpeng launched the full-stack self-developed NGP high-speed autonomous navigation and driving. At that time, Xpeng became the second company in the world and the first in China to develop autonomous driving technology based on an open computing platform, following Tesla.
In April 2020, Xpeng partnered with Nvidia, whose Drive OS enabled Xpeng’s patented autonomous driving software, Xpilot 3.0, to realize its full potential and continuously enhance its autonomous driving capabilities through OTA updates, By the end of 2022, the Xpeng G9 was equipped with two Nvidia Orin chips, achieving a computing power of 508 TOPS.
In the second half of 2021, Xpeng decided to independently develop driving assistance algorithms without relying on high-precision maps, leading to the creation of XNGP. It introduced a new perception architecture, XNet, enabling all-scenario intelligent driving assistance with just two Nvidia Orin X chips.
In November 2023, Xpeng fully launched its map-free urban navigation feature. By January 1 of 2024, the total number of cities covered by Xpeng XNGP reached 243, far ahead of the competitors.
From the second half of 2023, the ‘AI chauffeur’ feature became another strong asset for Xpeng’s intelligent driving. In January 2024, He Xiaopeng revealed that Xpeng has over 3,000 personnel on their AI R&D and data team, with an annual investment of C¥ 3.5bn.
At the heart of Xpeng’s technological advancement is the soon-to-be-integrated Turing chip. Developed by a team of over 200 specialists at the Shanghai R&D Center, this chip represents a major leap in Xpeng’s autonomous driving capabilities. The Turing chip, designed for L4 autonomous driving, has a 40-core processor capable of running large models with up to 30 billion parameters.
The chip supports two independent ISPs, ensuring a robust vision even in challenging conditions such as nighttime driving, rain, or backlighting. This feature is crucial for maintaining safety and reliability in autonomous vehicles.
XPENG Sensing Strategy for NOA (Phase 1) – Starting with Lidar
On September 15, 2021, Xpeng launched the P5 electric sedan featuring two lidars (HAP from Livox) in its top variant. Subsequent models, including the P7 facelift (P7i), G6, G9, and X9, also incorporated two lidars.
Supported by HAP lidar, the traditional ADAS system can be more effective in dealing with scenarios such as highway pilot, close cut-in vehicles in urban traffic jam, night driving, adaptive cruise on curves, and park-by-memory, so users can enjoy a better functional experience with ease and security.
Xpeng uses a combination of lidar, radar, and a camera for driver assistance. Xinzhou Wu says, “Lidar will provide the 3D drivable space and precise depth estimation to small moving obstacles even like kids and pets, and obviously, other pedestrians and the motorbikes which are a nightmare for anybody who’s working on driving”, the radar will detect the vehicles with the ability to detect the speed of the object and its location. The camera will provide the vehicle with basic semantic information as stated.
Details of the algorithm
Details of Livox lidar – Model HAP (T1) / HAP (TX)
Wavelength: 905 nm
Laser Safety Class: 1 (IEC60825-1:2014)(Eye Safety)
Detection Range (@100 klx): 150 m @ 10% reflectivity
FOV: 120°V × 25°H
Distance Random Error: (1σ @ 20m) < 2 cm
Angular Random Error:（1σ ） < 0.1º
Beam Divergence: 0.28°V × 0.03°H
Angular Resolution @ROI: 0.23°V × 0.18°H
Point Rate: 452,000 points/s (first or strongest return)
Data Port HAP (T1) : 100BASE-T1 Automotive Ethernet / HAP (TX) : 100BASE-TX Ethernet
Data synchronization: gPTP
False Alarm Rate (@ 100 klx): < 0.01%
Operating Temperature: -40 to 85 °C
Power: 12 W typical (Startup: 26W)
Power Supply Voltage Range: 9 to 18 V DC
Noise: 40cm omnidirectional <45 dB (A)
IP Rating: IP67
Dimensions: 105 × 131.6 × 65 mm
Weight: ~1,120 g
XPeng Sensing Strategy for NOA (Phase2) – Now launching some car models without Lidar, some with Lidars
Tesla’s assisted driving system has historically avoided lidar, which is expensive to implement. In contrast, most Chinese EV manufacturers, excluding the Baidu-backed Jiyue, have used lidar in their advanced smart driving software.
Xpeng P7+ launching without lidar – October, 2024
Xpeng’s latest model is the first of theirs to return to the vision-based approach to self-driving ability, leaving out the lidar. the system has 26 sensors: 11 cameras, three millimeter-wave radars, and 12 ultrasonic sensors. That camera total includes a driver monitoring system. The front-view binocular and rear-view cameras use the industry’s first single-pixel Lofic architecture. This helps the system see further where lighting conditions are far from optimal, such as at night. Processing all the data are two Nvidia Orin X chips with a total of 508 TOPS computing power. Pre-sales began at C¥ 209,800 ($29,450).
Xpeng’s regulatory filings share plans for facelifts to its G6 and G9 EVs, each abandoning lidar – December , 2024
According to regulatory filings in China, Xpeng plans upgrades to at least two current models. The filings detail facelifts to the Xpeng G6 and G9 SUVs and hint at the automaker’s strategy of droppinglidar sensors in favor of just cameras.
Xpilot 4.0 – full scenario advanced driver assistance
The G9 is Xpeng’s first model equipped with Xpilot 4.0, the company’s in-house proprietary advanced driver assistance system. Xpilot 4.0 will be the first to deliver full-scenario assisted ‘smart’ driving from vehicle start-up to parking, in an important step towards achieving full autonomous driving in the future.
Xpilot 4.0 is built on a new-generation hardware platform, with 508 TOPS ECU computing power supported by two Nvidia Drive Orin autonomous driving SoC units, 8-megapixel front binocular camera and 2.9-megapixel sideview cameras (covering front, rear, left and right view), and a highly integrated and expandable domain controller.
2024 – Xpeng keeps two lidars when launching a new variant of their X9 MPV – Sept. 25, 2024
Xpeng Motors launched a new advanced version of their X9 MPV, featuring lidar and XNGP ADAS.
When it launched, the X9 variants cost was between C¥ 359,800 ($51,213) and C¥ 419,800 ($59,750) respectively. Now a new variant of the X9 MPV called the 610 Max offers advanced driving features and lidar, along with a higher price. The decision to add lidar is an interesting one, as Xpeng recently dropped lidar and went with just cameras on their new P7+ BEV.
Details of Intelligent driving features at Xpeng (Car model G9)
The intelligent driving functions of Xpeng G9 include the following main functions:
- ACC adaptive cruise: This function can be activated by turning the gear lever down once, which can help maintain the distance from the vehicle in front, and at the same time, the distance and speed with the vehicle in front can be set by the button on the left side of the steering wheel, and the function can be exited by pressing the brake pedal.
- LCC Lane Centering Assist: When activated, the system can assist the driver in controlling the steering wheel and continuously centering the vehicle in the current lane. This feature is suitable for dry road conditions on highways with clear lane markings and should never be used on city streets.
- NGP intelligent navigation assisted driving: In high-speed road conditions, the gear lever can be quickly toggled downward, and the intelligent navigation assisted driving can be realized based on the navigation to reach the destination. In this mode, intelligent lane change overtaking can be realized, and when the speed of the vehicle in front is too slow, the car will automatically turn its turn signal to start changing lanes.
- XNGP full-scene intelligent assisted driving: This is a full-scene intelligent assisted driving system launched by Xpeng Motors, which can realize functions such as intelligent lane change, overtaking, and on/off ramps, and automatically adjust the speed and distance of vehicles according to road conditions and traffic conditions. In addition, XNGP can provide real-time traffic information and navigation guidance to help drivers better plan their routes.
- ‘AI’ driving: This function supports point-to-point customized intelligent driving routes within 100 kilometers, and up to 10 commonly used routes. After arriving at the navigation destination, you can open the route management on the route button on the SR page, click Customize to save, and wait for the review to be completed within 24 hours before creating a successful route.
Xpeng’s fully upgraded XNGP system emphasizes “nationwide usability” and delivers high-level autonomous driving experiences powered by an end-to-end large model and the AI Eagle Eye vision system.
By July 2024, the XNGP system had achieved a full coverage across all Chinese cities, marking a significant milestone in its evolution from being merely “usable” to truly “user-friendly.” Xpeng aims to complete a “door-to-door” autonomous driving solution, covering highways, urban roads, and parking, by the end of 2024.
In August 2024, Xpeng unveiled their ‘AI’ Eagle Eye vision solution. This system upgrades front, rear, and side cameras while eliminating the need for lidar, all while maintaining comparable intelligent driving performance. The system improves perception distance, recognition speed, and adapts to various lighting conditions, significantly enhancing the precision and safety of intelligent driving.
With bi-weekly updates to its intelligent driving system and a strong focus on nationwide deployment, including the neural network XNet, the control model XPlanner, language model XBrain, Xpeng is at the forefront of China’s autonomous driving revolution.