On December 6-7, 2023, the GTM2023 (6th) Global Technology Mobility Conference, hosted by Yiou Automobile, a think tank and innovation service platform for the technology travel industry, was successfully held in Beijing. With the theme of "Crossing the Cycle and Sailing Away from the Wind", this conference focused on the development and evolution of China's intelligent electric vehicle industry, and joined hands with executives from OEMs, first-chain enterprises, technology companies, scientific research and investment institutions to discuss the prospects of the industry, presenting an annual event of the travel industry with new technologies and new ideas. Mr. Huang Chi, Vice President of Momenta, shared with us the theme of "Scalable Driverless Critical Path".
In ".Forum of Scientific Entrepreneurs & Young EntrepreneursMr. Huang Chi, Vice President of Momenta, delivered a speech entitled "Scalable Driverless Critical Path". He believes that "one flywheel" is a data-driven AI flywheel, which is inseparable from the "two legs" of mass production intelligent driving and L4 completely unmanned driving, and mass production intelligent driving provides massive data to help train L4 algorithmsThe completely unmanned driving of L4 has raised the upper limit of the development of mass production intelligent driving technology, continuously improved the function and performance of mass production intelligent driving, and made alternating progress on "two legs", and finally realized large-scale unmanned driving.
The following is a sharing record for the reference of industry professionals:
Huang Chi: Good morning, everyone!I am Huang Chi from Momenta, and I am very happy to be invited by Mr. Yang to have the opportunity to communicate with you. Let me briefly share with you some of Momenta's thoughts and practices in realizing large-scale unmanned driving.
Let's start with Momenta. Momenta is an artificial intelligence company founded in 2016, and the core founders of the company at that time were scientists who were committed to AI research, and we were looking forward to bringing AI technology to our lives. So at the beginning of the company's establishment, we set a mission:"better ai, better life."Create a better life with better artificial intelligence.
With this mission in mind, we searched across the industry and found that the most valuable and meaningful thing in AI applications in the next 10-20 years is autonomous driving. After deciding to do autonomous driving, we set a goal to achieve large-scale autonomous driving. The most important point is that we need to make autonomous driving ten times safer than human driving through AI technology, so that we can realize our three visions in the next ten years: saving millions of lives in ten years, liberating 100% time in ten years, and doubling the efficiency of logistics and travel in ten years.
In order to achieve this goal, we have developed a very figurative strategy called "one flywheel, two legs", which is the key path that we believe can truly achieve large-scale autonomous driving.
This path is both a technical belief and a very pragmatic business consideration. The challenges of large-scale autonomous driving include millions of problems. In order to effectively solve these problems, we foresee the need to collect and analyze hundreds of billions of kilometers of driving data. So, we have two pragmatic considerations:
First, we chose a data-driven approach to replace the rule-driven approach that was prevalent in the industry at the time. In the early development stage of autonomous driving, the previous problem can be solved efficiently by rules, but as the number of problems increases, the rules-driven approach will face the problem of increasing costs and unsustainability. We clearly understand that only data-driven can support large-scale autonomous driving in the closed loop of business.
Second, faced with the need for massive amounts of data for algorithm training, as a start-up, we realized that hundreds to thousands of R&D vehicles were not enough. In order to achieve data collection of 100 billion kilometers, mass production must be done.
Based on these two points, we proposed a strategy of "one flywheel, two legs". "One Flywheel" is a data-driven AI flywheel, with three core elements: solving problems in a data-driven way, training algorithms through massive mass production data, and filtering and processing massive data with a closed-loop automation tool chain. This is inseparable from the "two legs" of mass production intelligent driving and L4 completely unmanned driving, and mass production intelligent driving provides massive data to help train L4 algorithmsThe completely unmanned driving of L4 provides the upper limit of technological development for mass production intelligent driving, continuously improves the function and performance of mass production intelligent driving, and makes alternating progress on "two legs", and finally realizes large-scale unmanned driving.
At this point, you may think that this is an obvious truth, but when our company was first founded, it was a strategy that was not optimistic, because it was already difficult to do one thing well. Some industry experts say that as a start-up company wants to become the leading business of the main engine factory, negotiate with the international main engine factory for five years, and then do it for another five years, and then the product can be market-oriented after ten years, and the domestic main engine factory will take five or six years even if it is faster, but as a start-up, there is no time and capital for us to use five or six years to verify the commercialization of the product.
But we look at this technology path again today, and find that it has become the mainstream of the industry, and more importantly, we see the acceptance of intelligent driving by users in the market today, which verifies our hypothesis at that time. I remember last year there were about 200,000 cars equipped with advanced driver assistance, for example, if you take high-speed navigation assistance as an example, this year it will be about 80-1 million vehicles, and next year we estimate that it will exceed 2 million units. What is this concept?New energy vehicles are now about 7 million units a year, and when a product can break through the penetration rate of 15% and be accepted by 15% of users, it will change from a niche product to a mass product.
At the same time, we have also seen that a year or two ago, many users would think that intelligent driving was just a nice to have, but since this year, we have seen more and more users willing to pay for high-end intelligent driving. The experience of high-end intelligent driving products has reached the user's usage threshold. The market for high-end intelligent driving has begun to detonate, which will further accelerate the accumulation of intelligent driving data and the iteration of algorithms, promote the development of the industry, and help realize large-scale unmanned driving.
In order to achieve our goals, we also have three very important tools at the R&D level, which we call "three treasures". The first treasure is the Momenta Framework, which ensures that L2 and L4 products are consistent on the underlying algorithm framework. In this way, the data between the two can be connected, and L2's data can be used to train L4's algorithm, and L4's algorithm optimization can also feed back L2's product iteration.
The second is Momenta Adaptor, as the needs of different OEMs need to adapt to different sensor platforms, computing platforms and assembly configurations, Momenta Adaptor, as a set of tool systems, its main role is to decouple the algorithm layer from the signal input layer to simplify the development process of each module. Developers can quickly adapt to new sensor data without modifying the core algorithm, and finally unify into a consistent signal input to the algorithm layer, so as to achieve efficient mass production.
The third is the Momenta Box, a development kit based on multiple production projects that solves the problem of software running on unstable hardware and delivering it in a timely manner. Typically, the product development process of OEMs is based on hardware development time. Momenta Box enables developers to develop software in parallel on unstable hardware, and once the hardware version is determined, they can quickly switch to the stable version, ensuring that both hardware and software meet production standards.
Let's take a quick look at our two-legged products. Momenta Pilot is a "full-scene, more continuous" intelligent driving product, a highly intelligent driving solution for mass-produced passenger cars, which can realize a variety of intelligent driving functions of different degrees and provide an end-to-end intelligent driving experience covering all scenarios.
Momenta Self Driving (Right Leg MSD) is a fully driverless autonomous driving solution designed for Level 4 and above, which can achieve full autonomous driving in personal vehicles and unmanned taxis. In fact, in addition to robotaxi, the demand for autonomous driving of high-end personal passenger cars is also becoming increasingly strong, and we are now continuing to negotiate with many international OEMs to promote this work.
I just said so much, Momenta company now only has more than 1,000 people, we made the first model when it was more than 1,000 people, today we do dozens of models we still have more than 1,000 people, that is, we talked about earlier "one flywheel, two legs, three treasures" to ensure that we are now efficient.
At present, we not only serve the local business of domestic OEMs, but also serve the overseas expansion of domestic OEMs and the global projects of international OEMs.
In the seven years of the company's development, we have five important strategic investors, including OEMs and Tier1. Making the customer a shareholder is an illustrative issue, because doing the project is the best due diligence. Autonomous driving is actually a very long-term work, what we deliver today is just a starting point, and there will be a continuous iterative process in the future, so it is not a one-shot deal, we need to reach a long-term strategic cooperation with OEMs, and continue to work together to make progress. We believe that to do a project, we must do it well, in order to strive for more cooperation opportunities, and finally reach a deeper cooperative relationship. I believe that with this concept, we will do a good job in products and serve customers, and we will also go more steadily.
That's all for today, thank you.