A Big Event for Jidu Auto, the Long-awaited Release of ROBO-01 Moon Exploration Edition
Authored by Roomy
Edited by Zhou Changxian
After a buildup of 605 days, Jidu Auto, which had been releasing information in a dribs and drabs manner, finally made a big splash.
On October 27th, Jidu officially released their first auto robot, the ROBO-01 Moon Exploration Edition, for the price of RMB 3.998 million.
As soon as the price was announced, social media was abuzz with comments. Some people said, “it’s wishful thinking to sell 10,000 cars a month at this price,” while others said, “how can Jidu match Tesla when their car is so much more expensive?”
Despite the many doubters, the publicity effect was good. According to an insider at Jidu, “we don’t need to fire the public relations department anymore.”
After the press conference, Jidu CEO Xia Yiping responded to some of the criticisms in a media interview. After more than an hour of communication, we learned that this engineer-turned-CEO not only played a decisive role in defining the first car model, but also demonstrated his thinking and judgment on the future competition pattern of the intelligent auto industry and Jidu’s next steps in development.
“The Real Competition Has Just Begun”
We all know that timing is everything in the auto industry. So, has Jidu, which has been established for over 600 days but has yet to mass-produce and launch its first car model, lost the game against time?
Xia Yiping does not agree with this question. “To say that we’re late is a bit late, but I think it’s relatively late.” This “relative” is his view that the first half of the electric vehicle competition has become increasingly homogenized, while the competition in the second half of the intelligent industry is just beginning.
Xia Yiping believes it’s not too late to enter the market now, and he cites two reasons: first, the electric car market is still a growing market, and second, the first wave of electric cars has already leveled the foundations of many products.
Currently, intelligent technology has reached an unprecedented height. In order to gain an advantage in homogenization, Xia Yiping is very clear that there must be its own characteristics.
“What are we good at? It should be clear that our best is breakthroughs in software and intelligent technology.”
The so-called breakthrough is to achieve the top-level intelligent capabilities of the ROBO-01 lunar exploration limited edition. “What makes us different is that our intelligent technology is standard throughout the system, and even the entry-level model is equipped with full laser radar.” In addition, it is equipped with the Qualcomm 8295 chip, dual Orin X intelligent driving chips, advanced intelligent driving, point-to-point navigation assistance PPA, global offline voice, cockpit and driving fusion, and other intelligent technology.
Advanced intelligence cannot be separated from the support of technology. Jidu Auto has combined the technologies of two companies, Geely and Baidu. The SEA platform comes from Geely, while intelligent driving and intelligent networking come from Baidu.
Jidu’s self-developed advanced automatic driving intelligence architecture JET integrates the electronic and electrical architecture EEA and the SOA operating system, which can carry core functions of high-level automatic driving and AI capabilities, realizing full-domain integration.
“Our vehicle’s electronic and electrical architecture is the strongest in the industry,” Xia Yiping believes that in addition to hardware, to gain an advantage in intelligent capabilities, “software is a systemic ability, not just copying others.”
However, the ability to think and implement is not just a “pure software” concept. It also requires a full understanding of how future chip technology will develop, how sensors will be made, and which direction intelligent technology will evolve. This is a systemic ability that not everyone can possess.“`
“To create intelligent vehicles involves high bandwidth and software security. We have implemented double redundancy for our smart cabins and intelligent driving systems, which is a first in the industry and relies entirely on our technical expertise and strong chip control. Currently, Jidu Auto is considering the software layout for the iteration of our next-generation chips.
The Jidu ROBO-01 uses the Qualcomm 8295 chip, which is the first car-grade chip produced using a 5nm process with an AI computing power of 30TOPS. This achieves nearly an 8-fold performance improvement over the Qualcomm 8155 chip. Many of Jidu Auto’s software architectures are built around this chip, such as the ability for smart cabins to directly access any smart driving domain, which is not present in many traditional vehicles.
However, Jidu Auto must also acknowledge a reality: a complete set of technologies, such as redundant automatic driving, multiple laser radars, high-precision maps, and intelligent cabins, are not new in electric cars.
Although Jidu Auto has always emphasized their competitive edge in offering more intelligent products to users and transforming the industry through technology to provide a unique customer experience, Mr. Xia Yiping, the CEO of Jidu Auto, has repeatedly emphasized that the core of intelligent competition is still the user experience.
The challenge lies in how to make users experience Jidu’s “futuristic” and “empathetic” vision, and this requires great skill. However, Jidu Auto does not simply innovate for the sake of innovation.
“We do not create products simply for the sake of innovation, but rather to consider whether they bring true value to users, improve user experience, and solve their pain points. Therefore, educating users is not as challenging,” explained Mr. Xia Yiping, who views “solving users’ pain points as innovation” as a fundamental product philosophy inherited from his internet background.”
“`”We have made many bold breakthroughs and innovations in the car, but behind these things, there are very high requirements for software. With confidence in our software capabilities as the basis, Xia Yiping has put a lot of energy into product design. He hopes that in terms of design, at least what you have, I will definitely have, but what you don’t have, I have.
The Jidu ROBO-01 lunar exploration limited edition has eliminated most of the physical controls such as exterior door handles and central control buttons, and is the first in the country to achieve mass production of the U-shaped steering wheel design that highly restores the concept design… These innovations that highly restore the concept design are somewhat bold and stimulating.
Why is speech recognition important?
On the Jidu ROBO-01 lunar exploration limited edition, in addition to the redundancy of the intelligent cabin and intelligent driving, Jidu also demonstrated for the first time the production of a 3D intelligent driving map and global offline speech recognition.
As we all know, high-precision maps are a necessity for high-level autonomous driving. In terms of importance, it is no lower than LiDAR. Some believe that high-precision maps and LiDAR are the two crutches of intelligent driving – one responsible for opening the way and the other for obstacle avoidance.
In the competition in the field of autonomous driving in the past two years, the competition for LiDAR is the most direct, with car companies not only burying multiple LiDARs, but also placing great importance on “how many LiDARs there are”. Competition for high-precision maps, compared to the naked competition for LiDAR, seems to be a bit cooler.
Regarding the future of high-precision maps, there is generally an industry view of “transition”.
He XPeng said, “For autonomous driving, high-precision maps must be a transition step, and cloud integration is only an assistance. True autonomous driving must be able to drive in all scenarios.” Yu Chengdong also stated, “High-precision maps are still being used for autonomous driving, but future development should not rely on high-precision maps and vehicle-road coordination.”
Jidu has its own opinion on high-precision maps. Xia Yiping said, “It is not realistic to completely get rid of high-definition maps in the short term.” However, he pointed out that in terms of long-term technology introduction, automatic driving is doing remote recognition and high-definition maps will certainly not be required in the future.
But technology and algorithms need a process of iteration. Before achieving fully autonomous driving, both LiDAR and high-precision maps are indispensable. Therefore, Jidu is also continuously working on high-precision maps and maintaining a high level, mainly thanks to Baidu’s technology.
Xia Yiping believes that “Baidu has had its own supercomputer platform for many years, and many models with large computing power need to use the supercomputer platform. This is one of our advantages.” For Baidu, especially an AI company, entering the intelligent automotive field requires a completely different core competitiveness.
The fully integrated 3D intelligent driving map displayed by Jidu provides clearer visualization, making lane-level navigation clear at a glance. It achieves the fusion of static map navigation and dynamic perception data, allowing vehicle robots to perceive and navigate on a three-dimensional map.
Xia Yiping said that Jidu needs to satisfy three points: natural communication, free movement, and self-growth. The reason why more innovation is maintained on the map is that Jidu’s vehicles need to meet these three points.
With strong software capabilities and thinking about future growth, Jidu has the foundation for “self-growth”, and the 3D intelligent driving map provides support for “free movement”. So how can “natural communication” be satisfied?To know, natural communication is the most direct and often overlooked aspect of the intelligent cabin experience. Therefore, Xia Yiping was asked why Jidu Automotive spent so much effort on voice recognition.
“Perhaps people underestimate the importance of voice recognition. The significance of solving this issue is no less than that of autonomous driving, and the complexity is no less either.” Xia Yiping hopes that every car produced by Jidu, in terms of voice recognition, should not only involve simple commands but also truly achieve a very natural language communication.
It sounds easy to say, but it’s not an easy task to accomplish. It’s easy to recognize a simple statement like “close the window and open the window” by running a script locally. However, recognizing natural language involves running to the cloud for recognition and coming back around. It is also affected by network issues, as in, “go to the underground garage and have a look.”
“The most important thing in ensuring voice interaction experience is to let a robot communicate with people naturally.” Xia Yiping said that Jidu’s goal is very simple, to achieve “visible and capable of speaking” and “touchable and controllable”. Any robot car that goes out with people needs to recognize the surrounding environment. This is also the DNA that Jidu Automotive has always emphasized: futurism, robotization, and empathy.
Thanks to the powerful computing power of Qualcomm 8295, Jidu has achieved full localization of voice AI algorithms, with a voice recognition speed of up to 500ms on all-end sides and an end-to-end voice response speed of under 700ms, completely eliminating the reliance on network signals. Even in weak network or offline scenarios such as tunnels, underground garages, and mountainous areas, smooth interaction is possible.
Jidu hopes to create a new type of car robot, a robot car that “understands people.” If this concept is not explained well, it may come across as “suspended.” Xia Yiping is often asked if he is worried that this concept is not grounded or difficult for users to understand.
“To be honest, I do have worries, and I must be frank.”However, the accumulation of Baidu in the field of AI, including the accumulation in various speech technologies over the past decade, has given Jidu some confidence in autonomous driving. Therefore, we dared to come out and proclaim that we are truly creating a product that is like a robot experience.
This is also the starting point for Jidu’s emphasis on voice.
The Purpose of “Stacking”
From Jidu’s attitude towards 3D global maps and emphasis on voice, we can also see Jidu’s basic strategy of intelligence, that is, the so-called “stacking,” which is not cheap.
In an interview, Xia Yiping was directly asked why the Jidu ROBO-01 lunar exploration limited edition had seven 8 million cameras, and whether the use of such high-precision cameras significantly increased the cost.
The implication was, is it necessary?
Xia Yiping answered frankly, “It is relatively simple for us to do autonomous driving. We need to see what is in the scene library to achieve what we want. If we want to draw a picture, the right side is all the abilities that need to be achieved, and the left side is the hardware configuration.”
As for why so much hardware is needed, Xia Yiping admitted, “As a start-up company, Jidu still hopes that the system will be more secure at the beginning. We are still more aggressive in promoting products, so we make some redundant configurations on the overall hardware.”
Currently, in the driving scheme, the Jidu ROBO-01 lunar exploration limited edition is the only product in the industry that uses two sets of intelligent driving perceptual schemes, “vision + LiDAR,” which run independently, back each other up, and cooperate with each other.
Why go to such great lengths to do autonomous driving? Xia Yiping said that it was for the same reason as emphasizing voice, “It’s not that we’re like others, just putting an L2 on a low-end model. We’re here to offer it as a standard across the board.” Even the entry-level version comes standard with full LiDAR.In terms of the use of LiDAR, Jidu has also made some compromises: it is possible to do without LiDAR, but the capability is greatly reduced. Once LiDAR is added, Jidu can handle low-speed scenarios, automated parking, and many special autonomous driving scenarios.
At present, Jidu’s primary goal is to achieve out-of-the-box deployment and the deliverability of a product that users can use. That’s why many of Jidu’s technical capabilities are built from the starting point of ROBO-taxi, approaching L4.
In fact, Xia Yiping, who comes from a technical background, can’t separate his thinking from Jidu’s auto product logic and DNA: futurism, robotics, and empathy.
“We won’t do gimmicky things, we need to make great changes to the user experience,” Xia said. He believes that the competition for future intelligent cars will shift from traditional manufacturing capabilities to intelligent capabilities, which is a trend that is happening right now.
“Traditional capabilities are not unimportant, they are the foundation, and we need to have them because others have them as well. However, I believe that the core competitiveness will gradually shift to the ability to control computing power, algorithms, and chips.”
Like ROBOTaxi, the most beautiful thing about AI is that as long as it has enough capability, it can grow on its own. “Good products are still the same, user education costs are very low. When cars become more intelligent, they can help you do things on their own, and you don’t have to worry, this is actually an understanding of technology.”
Currently, Jidu’s autonomous driving technology has a iteration cycle of “every week”.
Summer Yiping revealed that “we are a team integrating with Baidu, which is why we were able to achieve many autonomous driving scenes early on. We integrated Baidu’s capabilities and as we deliver more products, we will achieve more scenes. Many of the algorithms in the backend are shared and our goal is also the same, which is to promote the implementation of higher-level autonomous driving capabilities.”
Of course, this does not mean that podding only means “adding ingredients”. From the perspective of long-term iteration, more is not necessarily better, it is important to find what is suitable.
He believes that “this is an iterative process. After the entire system is online and in mass production, the number of sensors may be reduced. In the end, it is not about who can buy the best hardware, but about continuously optimizing the software capabilities and improving system efficiency.”
But this is a process, it cannot be achieved overnight.
Closing Remarks:
Podding’s strategy is very clear. Yiping’s thinking is also clear, “Podding is still a startup. In the 0 to 1 process, our approach differs from many mature companies because we need to retain users in a timely manner.”
He believes that in the early stages of building a brand, it is necessary to convey one’s own ideas. For Podding, maximizing skills is a strategy. “We hope to compete in a relatively high-end market and our focus is intelligent vehicle models that are premised on intelligence.”
As the Podding ROBO-01 lunar exploration limited edition is launched, Yiping has not relaxed. His focus is also changing from product definition and technical research to the establishment of future sales and service systems.
“I tell all the teams that just manufacturing cars does not guarantee success. Manufacturing the car is not the only important thing anymore.” Yiping said that in October of this year he has already set the entire vehicle order quantity for next year and the entire production and sales channels will gradually run smoothly.
“These are important tasks for us in the future and will only become increasingly difficult.” However, he is confident in the product and will appropriately add some advance orders.
“My goal is to sell 10,000 cars every month.” Yiping said confidently.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.