Smart Driving "Methodology" for Ideal Cars

"The essence of doing a good job of autonomous driving is actually the effective operation of the data closed loop." Lang Xianpeng, vice president of autonomous driving of Ideal Automobile, delivered a wonderful speech at the 2021 Annual Meeting of Gaogong Intelligent Vehicles, saying that this is also an important reason why Ideal Automobile chooses the intelligent driving function standard of all models.

Ideal Automotive Vice President of Autonomous Driving Lang Xianpeng

At present, there is still a long way to go to achieve L4 level autonomous driving, but L2 level intelligent assisted driving is accelerating penetration and gradually becoming a standard function of intelligent driving cars.

According to the data of the "Gaogong Intelligent Automobile Research Institute", from January to November 2021, the number of L2-level ADAS in the front of China's new cars (joint venture + independent brands) was 3.3851 million vehicles, an increase of 78.33% over the same period last year, only lagging behind the L1 capacity of about 70,000 units.

Ideal Car believes that the realization of L4 autonomous driving requires the application of a large amount of data and the gradual accumulation of iterations. The market penetration rate of L2-level ADAS has reached the critical point of crossing the gap, and has become an important foundation for major car companies and suppliers to sprint for high-end automatic driving.

Safety and standard equipment are therefore the primary targets of the ideal car intelligent driving product. Lang Xianpeng said that as of November 2021, the total mileage of ideal car assisted driving has reached more than 160 million kilometers, and the number of AEB triggers has exceeded 60,000 times.

First, the double closed loop of perception and regulation is the key

As we all know, the core of autonomous driving is algorithms and data. Only by continuously recovering the real driving scene data through the sensors equipped with the vehicle in motion can players feed back the iterative upgrading of intelligent driving technology and algorithms.

The industry generally believes that with the continuous loading of ADAS functions, the boundaries of intelligent driver assistance systems can be continuously expanded, which is also the most critical step in the large-scale landing process of high-end intelligent driving technology.

Lang Xianpeng said that data is the key road to high-end intelligent driving, but to achieve full-stack data-driven intelligent driving, the first thing to achieve is the double closed loop of perception and regulation technology.

On the one hand, the rhythm, planning and technology development of intelligent driving product launches are an "interlocking" process. Among them, the algorithms, data, hardware and software of intelligent driving technology are also an interdependent and mutually influencing "interlocking" relationship.

On the other hand, the essence of the competition for autonomous driving is the effective operation of the data closed loop, but the premise is that it is necessary to obtain high-quality driving scenario data, including intelligent driving data itself, high-precision maps and OTA data. This requires enterprises to understand the user's needs for intelligent driving scenarios, and do a good job in intelligent driving perception and regulation and other technologies. Only in this way will users rely more on intelligent driving and increase the frequency of use of intelligent driving functions.

Ideal Car believes that only when the intelligent driving function is equipped with the standard in the model, can the real driving scene data be better obtained for the automatic driving algorithm to train.

"If a model sells 100,000 vehicles per year, but the optional rate of smart driving functions is only 10%, that is, only 10,000 cars actually return intelligent driving data every year." Lang Xianpeng said that this is an important reason why the ideal car chooses the intelligent driving function of the whole series of models as standard.

In addition, although L2-level ADAS has maintained a rapid growth trend, the competition in the entire market has gradually become fierce. Lang Xianpeng said that the competition of intelligent driving should not only focus on physical links, but also pay attention to data processing and software algorithm links, and the most critical competitiveness lies in the overall strength of software and hardware.

Second, the characteristic intelligent driving road of the ideal car

At present, the intelligent driving industry is also facing pain points and challenges such as safety, such as the industry's solution to safety problems is still different, and safety standards have not been unified.

However, ideal car has established a set of perfect ideas and guidelines for intelligent driving, including the demand cognition of intelligent driving, product standards, intelligent driving product development and technology research and development process.

First of all, in terms of cognition, ideal cars believe that if they want to do a good job of autonomous driving, they must use the exploration of the unknown and open ideas to do automatic driving, and set a long-term goal. Among them, the primary goal of intelligent driving of the ideal car is safety and standard equipment.

The second is the process aspect, that is, the IPD process traction of product development and technology research and development planning, the ideal car will first through full market research and insight, in-depth understanding of user needs, so as to make a new value for users to bring new value products.

Lang Xianpeng said that the product focus of the ideal car has always been the user value, that is, the product strength and product quality are put in the first place, and each intelligent driving iteration upgrade will be upgraded according to user discussion, feedback and other information.

For example, on December 6, 2021, Ideal Auto began pushing OTA version 3.0, and the new version added many features. Among them, 2020 users add "visual parking" function, and 2021 users can open the "navigation assisted driving" function.

It is understood that in the development process of NOA, Ideal Auto has completed more than 1 million kilometers of road tests, collected through actual tests, and also updated and corrected many inaccurate information in autoNavi and high-precision maps. This set of closed-loop, engineered data intelligence capabilities is the core of the ideal NOA capability.

The next is technology, ideal car through market research, in-depth understanding of what kind of technology the market needs, in order to create a sensor input and semantic space, semantic space and control signal of the dual closed-loop data drive.

Finally, there is the organizational aspect, which mainly comes from the cognition at the top. Ideal Car believes that autonomous driving is an open and uncontrollable technology that requires the creation of a co-creation, enabling organization and business approach.

Based on such a set of product ideas and processes, from the standard intelligent assisted driving of the whole series to the full-stack self-developed ideal AD assisted driving system, the ideal automobile intelligent driving research and development team has embarked on a characteristic path.

According to the data of the "Gaogong Intelligent Vehicle Research Institute", from January to November 2021, Ideal Automobile, as a representative of China's new local car manufacturing forces, entered the top ten of L2 ADAS front-loading standard capacity for the first time, accounting for 2.28% of the share.

"From vehicle-side perception, planning, control decision-making to data-side annotation processing, training and simulation, and vehicle adaptation, the ideal car has achieved full-stack self-research." Lang Xianpeng said that these will become an important foundation for the ideal car to move towards the next era of intelligent driving.