Guangzhou Auto Show: Xiaomi, Li Auto, Zeekr, and Xpeng jointly promote end-to-end

On November 15th, today's Guangzhou Auto Show became an arena for intelligent driving technology. Four new energy vehicle brands successively released their latest end-to-end intelligent driving technologies. Xiaomi, Li Auto, Zeekr, and Xpeng each showcased their intelligent driving solutions with large models and full scene coverage as the core.

 

Xiaomi Auto: Launch "Super Intelligent Driving" Internal Test

 

Xiaomi founder Lei Jun announced that "Xiaomi Hyper Autonomous Driving" (HAD) will start targeted internal testing on November 16th and launch the Pioneer version at the end of December. In terms of end-to-end capability, it can achieve intelligent driving experience for driving and parking in all scenarios.

 

The system is based on the accumulation of graph free NOA technology and user driving data training end-to-end technology, combined with VLM visual language large model, which makes the driving trajectory smoother and more natural, reduces sudden acceleration and braking, and can intelligently predict and plan the best path and make reasonable decisions in advance.

 

In addition, the vehicle can also start the parking space and use it immediately upon boarding, with stronger game ability, flexible response to oncoming vehicles, pedestrians, static obstacles, and the ability to pass through ETC and gate checkpoints.

In order to ensure driving safety, Xiaomi's end-to-end model adopts a free multi-dimensional evaluation mechanism, conducts real-time safety assessment, and embeds the safety bottom line into the genes. The system responds faster to complex road conditions such as confusion between two wheelers and pedestrians.

 

In addition, VLM technology can help vehicles analyze road conditions in real time, such as bumpy roads, construction scenes, and T-shaped intersections, and provide corresponding prompts for drivers.

 

Li Auto car: promoting "parking space to parking space" intelligent driving

 

Li Auto announced at the Guangzhou Auto Show that its end-to-end+VLM feature has officially pushed parking space to parking space to a group of ten thousand people, and plans to fully push it with OTA 6.5 within November, claiming to be the "world's first full push".

The new generation intelligent driving technology architecture of Ideal Automobile has added functions such as parking to parking intelligent driving, R-MAI backward false acceleration suppression, R-AEB backward automatic emergency braking, and handheld car infotainment system.

 

Fan Haoyu, Senior Vice President of Ideal Automotive Product Department, stated that based on end-to-end+VLM parking to parking function, the push of test drive cars for 10000 person groups and nationwide stores has been launched. By the end of November, all AD Max users can enjoy supervised intelligent driving from parking to parking.

 

Zeekr Auto: Release of the Vast Intelligent Driving 2.0 System

 

Zeekr Motors brings the vast intelligent driving 2.0 end-to-end Plus. The first fully self-developed architecture introduces Zeekr's first digital foresight network, which has road foresight, behavior foresight, and spatiotemporal foresight capabilities. It can cope with various road conditions, enabling the intelligent driving system to leap from "old drivers" to "local old drivers". With the advantage of full stack self-developed architecture, it can achieve efficient iteration, ensure safety and reliability, and bring users a more human like and worry free experience.

 

The digital foresight network pioneered by Extreme Krypton plays a key role in this architecture, possessing three core foresight abilities: road foresight, behavior foresight, and spatiotemporal foresight. Road foresight allows vehicles to cleverly navigate around potholes on the road while driving.

 

Behavioral foresight can detect and avoid dangerous emergencies such as "ghost probes" in a timely manner, while spatiotemporal foresight can effectively avoid the risk of road potholes. The end-to-end Plus is capable of handling the vastly different road conditions in different cities.

Its map free city NZP will be fully promoted to users nationwide by the end of 2024, and parking to parking navigation (D2D) will achieve nationwide full promotion in the second quarter of 2025.

 

This architecture can achieve efficient iteration, ensure safety and reliability, and is supported by the MLM multimodal big language model and secure bottom line model. The Extreme Krypton Map free City NZP can break free from high-precision map limitations and cover 100% of cities across the country, including 38000 townships, over 700000 administrative villages, and more than 2.6 million natural villages.

 

It is suitable for level 5 roads such as provincial roads, township roads, alleys, village roads, and even unpaved dirt roads, providing drivers with excellent driving experiences such as smooth detours, comfortable following, and efficient passage at complex intersections.

 

Zeekr is one of the first brands in the industry to support advanced assisted driving from parking space to parking space. Its D2D (Parking to Parking Navigation) function has three core features: full speed domain activation, full gate access, and intelligent parking with unlimited parking spaces in all scenarios.

 It is the first in the industry to achieve autonomous parking without pre registration, providing users with thoughtful parking to parking services. It is expected to start phased promotion in January 2025 and full volume promotion in the second quarter.

 

Xpeng Motors: Comprehensive Breakthrough in Intelligent Driving Solutions

 

Xiaopeng Motors announced at today's Guangzhou Auto Show that its "parking to parking" intelligent driving solution has launched its first test across the entire network. Of course, the software claims that NGP is used for urban roads and parking environments, rather than combining navigation and parking into one function. Xiaopeng Motors aspires to be the world's best intelligent driving solution, constantly exploring and improving.

 

Based on the Turing AI intelligent driving system, Xiaopeng Motors has become the first enterprise in the industry to use a set of intelligent driving software and an end-to-end large-scale model to achieve parking space to parking space. This solution uses a set of software logic to connect underground warehouses, gates, urban roads, and other scenarios. It also features light maps, light radar, and seamless route generation.

 

At the press conference, Xiaopeng Motors also announced the latest progress of its crowdfunding plan for the revitalization of old user chips. This plan is the first crowdfunding campaign for chip upgrades in the industry. Through chip upgrades, car owners will be able to enjoy more advanced intelligent cockpit functions and a more powerful intelligent driving experience. As of that day, the crowdfunding for the Xiaopeng P7 intelligent cockpit chip has been achieved, and the research and development project for the corresponding model has officially started. The crowdfunding activity is still ongoing, and we look forward to more users joining in.

 

Common features of end-to-end intelligent driving technology among Xiaomi, Li auto, Zeekr, and Xiaopeng:

1.End to end big model driven: All brands' technologies are centered around end-to-end big models, which generate overall intelligent driving logic through training data. From path planning to actual operations, all are directly completed by AI, reducing the limitations of traditional modular architectures.

 

2.The promotion of map free driving: Various brands are committed to promoting map free NOA (navigation assisted driving), relying on multimodal perception capabilities such as vision and radar to reduce dependence on high-precision maps and expand the adaptability of intelligent driving scenarios.

 

3.Full scene coverage from driving to parking: The technology of these brands emphasizes a complete experience of "parking space to parking space", covering complex roads, gates, parking lots and other scenarios, providing efficient and smooth driving and parking services.

 

4.The emphasis on safety: By introducing self supervised learning and real-time safety assessment mechanisms with large models, various brand technologies have improved safety in complex traffic environments, such as avoiding "sudden acceleration and braking" and handling "ghost probe" scenarios

 

All four car companies are committed to deepening end-to-end technology, using no image NOA, large models, and full scene coverage as technological breakthroughs to enhance users' driving convenience and safety.