This article is 18,000 words long and takes at least 60 minutes to read carefully.
“Before, when I was at my previous company, the team did a lot of analysis, and the ultimate conclusion was that the biggest factor affecting the landing of autonomous driving technology is — the scene.” At the end of last month, a system architect of an unmanned driving company at a large open-pit coal mine in Northwest China said while communicating with “Nine Chapters Smart Driving”.
In fact, as the exploration of commercialization goes deeper, more and more entrepreneurs on the subdivision racetracks of autonomous driving have formed the concept that “the understanding of scene will be the highest barrier”. Among many companies that the author has come into close contact with, the above-mentioned mining unmanned driving company is a typical representative that “surrounds scenes from top to bottom”.
In order to better understand the application scene, this mining unmanned driving company not only has a large number of technical personnel stationed at the mine for a long time, but even business and marketing students also stay at the mine for a long time. Recently, a few girls responsible for logistics also went to the mine for a week.
The founder of the company often jokes that we are the “poorest” unmanned driving company.
The more “poverty”, the more harvests. Combining the data disclosed by various competitors in the official WeChat public account and to the media, as of May 2021, the cumulative unmanned driving test mileage of this company is twice as much as the second place on the same track and four times as much as the third place.
Currently, the unmanned driving vehicles in the trial operation of this company at a certain mine in Northwest China are relatively stable in single-composition operation. Later, the company will focus on the hybrid dispatching algorithm of the two-composition and increasing the effort to remove safety personnel.
In order to speed up “solving” scenes, since January of this year, the company has also invested a considerable amount of resources in the traditional “manually driven” transportation business at the mine in Ordos, Inner Mongolia. Many peers and investors do not understand this behavior and even ridicule and question it. In fact, their “manually driven” business is to empower unmanned driving businesses.
Perhaps many reader friends who are familiar with the mining unmanned driving track have already guessed that the “company” we mentioned earlier is YikongIMa.
In early June last year, after 16 months of in-depth communication with YikongIMa CEO Watson, the author published a detailed introduction to YikongIMa in “Jianyue Car Evaluation” in an article titled “The Easiest Scene for Autonomous Driving Technology to Monetize: Open Pit Mining Transportation”. Friends who pay attention to the mining unmanned driving track should already know the content mentioned in it, and this article will not repeat it.From mid-June last year to the end of May this year, Jiuzhang Intelligent Driving (Jiuzhang) had in-depth exchanges with Watson and completed over 6 sessions, totaling more than 12 hours. In addition, Jiuzhang also had nearly 20 hours of discussions with nearly 10 members of the Watson team, including co-founder and chairman Zhang Lei, VP of Technology Lin Qiao, Chief Algorithm Officer, Chief Mining Engineer, Testing Engineer, and Operation Manager on site. These exchanges eventually led to the creation of this article.
Through this series of interviews, we gradually clarified the answers to the following questions:
- Why do traditional “manned” transportation?
- Why only use wide-body vehicles and not pursue large mining trucks?
- Why not develop a domain controller in-house?
- Is focusing on the simplest scenarios first considered to take a detour?
- Is there any concern about being marginalized by Robotaxi or companies that focus on trunk logistics?
- What is the most difficult barrier to competition for Yixiang Intelligent Driving that cannot be replicated?
During the interviews, the author also learned that there is a debate surrounding Yixiang Intelligent Driving. Most founders of other autonomous driving companies have an AI background, but Yixiang’s CEO has a gaming background. The question then arises: can someone without a technical background lead an autonomous driving company?
At the end of this article, the author will combine their own observations of the autonomous driving industry over the past few years, their learning about organizational culture, and their understanding of Watson to answer the question, “Can someone without a technical background really lead an autonomous driving company?”
Why pursue traditional “manned” transportation?
From January to now, Yixiang Intelligent Driving has invested a total of 20 vehicles for traditional transportation at the Donggedu coal mine in Ordos, and the fleet is expected to increase to 40 vehicles by the end of June. If traditional transportation runs smoothly this year, Yixiang plans to add an unmanned driving test site here.
However, both inside and outside the company, there are some people who do not understand why Yixiang is pursuing traditional transportation. Watson’s response to Jiuzhang Intelligent Driving was, “The more we work on traditional transportation, the more we realize the enormous help it provides to autonomous driving.”
According to discussions between Jiuzhang and Watson, as well as other members of the Yixiang team, the empowerment of traditional transportation to autonomous driving mainly involves the following aspects:
Helping the technical team understand the scenarios
Yixiang Intelligent Driving is not the first company to pursue traditional transportation in the autonomous driving field.
The earliest example may be Starsky, which went bankrupt last year. Even though the company is no longer active, some of its experiences are still used by other companies.In China, unmanned driving companies in the sanitation field, such as KuWo and Yu Wanzhi Jia, also engage in traditional sanitation services in order to understand the scenario. Among them, at Yu Wanzhi Jia alone, there are over 800 sanitation workers who only pay social security. In a communication in April of this year, Iris, the joint CEO of Yu Wanzhi Jia, gave an example and said, “Our sidewalk algorithm is actually written with reference to the lazy ways of sanitation workers. The coolest algorithm is still in the minds of sanitation workers.”
The mining scene has its own characteristics. If technicians do not understand these characteristics, it is difficult to do the algorithm well. The algorithm director of Easy Control Intelligence Driving gave an example: There is a task in the perception link to segment the ground out of the point cloud, but because the ground in the mine is not as flat as the asphalt road, segmentation is difficult.
“If you don’t have a concept and don’t understand what the normal working environment of unmanned vehicles will be like in the future, even if you have spent a lot of effort trying to make the algorithm bigger and more comprehensive, there will still be many urgent areas that will be missed. After traditional business is up and running, we have the opportunity to get scene data and use this data in our algorithm training.” said the algorithms director.
Of course, it should be pointed out that currently, due to the immaturity of traditional transportation services, Easy Control has not installed sensors on the vehicles yet. However, the work of installing sensors is being prepared and will be launched within a few months.
After communicating with the perception algorithm leader, Nine-Chapter Intelligence Driving learned that the environmental perception data provided by sensors such as LIDAR is only the tip of the iceberg. In addition, the driving behavior data of the vehicle provided by traditional transportation vehicles has greater value to the technical team for controlling algorithm and scheduling algorithm.
Next, Easy Control plans to install T-Box on some vehicles to collect driving data of the vehicles themselves. With the fuel consumption and failure data provided by traditional transportation vehicles, the technical team can know how to make simulation control algorithms more conducive to reducing fuel consumption and reducing failure rates.
For example, in the process of soil excavation, vehicles cannot always press down in the same position as last time, because if it is pressed for a long time, this place is prone to collapse. How to precisely control the position of soil excavation requires learning about the know-how from the traditional transportation team.
In terms of scheduling algorithms, currently, for issues such as where it is most appropriate to reverse when loading soil, how the vehicle matches the excavator, and how to do double-formation, the technical team can only explore how to do it through simulation or observation by filming videos. In fact, there are no accurate values. However, if the driving behavior data provided by traditional transportation vehicles is available, these issues can be solved more easily.
Currently, Easy Control has arranged engineers from unmanned driving teams such as simulation, maps, and scheduling to communicate fully with traditional business teams.
Reserve talents for large-scale landing of unmanned driving business.More than a decade ago, when Watson was still in the gaming industry, there were many market opportunities and sufficient funds, but he missed out on many chances because he could not find suitable talent to expand the business. This experience left Watson with a painful lesson: to start with the end in mind and make early preparations.
Currently, unmanned driving technology is still in the small-scale trial operation stage, with the vast majority of the team consisting of technical personnel and very few business and on-site management personnel. However, when unmanned driving technology matures enough to be widely used commercially, the company will need a large number of operating personnel. So, where will the workforce come from?
On the one hand, if we hire some people from outside as a stopgap measure, it’s questionable how reliable they will be. On the other hand, it’s not feasible to cultivate them in the unmanned driving team in advance, because the current unmanned driving team really doesn’t need such people. The addition of traditional transportation services can, therefore, help to reserve such talents for the commercialization of unmanned driving.
Watson said, “Some business talents have a deep accumulation in the traditional business field and will be very helpful for us to carry out large-scale operations in the future. However, we cannot find a suitable position in our unmanned driving business to place these people. If I have traditional business, I would allow them to spend 70% of their time doing traditional business and 30% doing unmanned driving business, as a buffer process.”
In the process of cooperation with the engineering company, Watson will carefully examine what kind of people are better and more likely to stay on site, and then recruit people based on this criterion. “Because we are an unmanned driving company, doing traditional business is only a transition, so it is easier to recruit business and operation talents than traditional engineering companies.”
As the core purpose of YiKong Intelligent is to empower unmanned driving business through traditional business, it is more willing to provide more attractive treatment and opportunities than other traditional engineering companies, both in terms of business and operational talent reserves and R&D scenario support. This mechanism is also more conducive to attracting and retaining talent.
Watson plans to find some undergraduates to stay for a year in the traditional business once it has grown slightly larger, and then transfer them to the unmanned driving team to do operations. In addition, they also plan to pick some outstanding people from the safety personnel to serve as the remote control operators and on-site management personnel of the unmanned driving fleet.
Watson said, “By doing traditional business, we can screen out people who are suitable for future operations in advance. Currently, our scale is small, and the cost of trial and error is relatively low. If we wait until the unmanned driving fleet is operating on a large scale and then find that we hired the wrong people, the cost of correcting the errors will be relatively high.” YiKong Chairman Zhang Lei and the chief engineer of the open-pit mine, Sun Qingshan, also hold the same view.
Reserving Business Resources for Unmanned Driving Business
Another benefit of doing traditional business is that it can enable YiKong Intelligent to integrate into the circle of mines in advance.Mine Circle is a closed circle. Even if your unmanned driving technology is very advanced, new soldiers may not easily recognize you. However, in the process of traditional transportation services, Yikong can gradually establish close relationships with mining enterprises and engineering companies. This will make it easier for unmanned driving businesses to enter in the future.
Currently, Yikong’s traditional transportation business is managed by a friend of Watson’s from many years ago. They worked together to make games before and then low-speed electric vehicles. His comprehensive ability and personal integrity have been verified.
The operation of open-pit mines involves several stages such as blasting, excavation, transportation, and drainage. Yikong is only responsible for excavation and transportation. Other stages such as refueling vehicles, road surface renovation and maintenance, and watering services are provided by traditional engineering companies.
I asked Watson, “Previously, the transportation was also done by engineering companies. Isn’t it taking their business away if you do transportation instead? Why do they still cooperate with you?”
Watson explained, “In fact, engineering companies are not willing to spend a lot of money to buy equipment with interest. Their core competitiveness is managing all kinds of relationships. We are not replacing engineering companies, but small subcontractors. For engineering companies, they used to manage many small subcontractors. After cooperating with us, they only need to manage us, which greatly reduces management difficulty.”
Overall, engineering companies welcome the entry of unmanned driving. Yikong first uses traditional business to integrate with the industry and build commercial resources, “pave the way” for unmanned driving to land, and then selects suitable scenarios for unmanned driving to directly switch to. This operation not only naturally increases the cooperation stickiness between mining companies, engineering companies, and Yikong, but also reduces the operational cost of the three parties through coordination.
Reserve Maintenance and Other Supporting Resources
In the next few years, the size of the unmanned driving fleet will gradually increase, but it will not be particularly large. This presents a great challenge for maintenance and repair – service providers are not willing to cooperate in onsite mining.
The vehicles used in Yikong’s traditional transportation business are also wide-body vehicles from Shanxi Tongli. Except for not using a wired chassis and not carrying an automatic driving kit, their structure and parts are similar to the models used by the unmanned driving team. This also means that the two parties have many similarities in the maintenance system.
At present, the maintenance system of the traditional transportation fleet is relatively mature (and can share resources with traditional engineering companies), and Yikong gradually introduces unmanned driving fleets, so there will not be many problems.
Advance the Bond Financing Channel
Yikong Intelligent Driving is adopting a heavy asset model. The total cost of one wide-body car plus sensor is over one million yuan. As the fleet grows, Yikong’s demand for funds will also increase. If the fleet size reaches 10,000 cars in six or seven years, each car costs 1 million yuan, which requires 10 billion yuan in funds.For companies whose valuations are rapidly rising, frequent equity financing is not cost-effective. If the valuation rises 50% in a year, the equity financing done a year ago is equivalent to an interest rate of 50%. In comparison, debt financing (usually with an interest rate of around 6%) is more cost-effective. However, unlike with equity financing, where several billion can be raised at once, acquiring debt financing from a bank requires a long time to accumulate credit.
Fortunately, YiKong’s traditional business not only has cash flow, but also has good profits. In fact, YiKong’s profit margin in intelligent driving is much higher than that of a traditional engineering company. The main reason is that:
When purchasing equipment, traditional engineering companies typically pay around 30% upfront, with the remaining amount being paid in installments at an annual interest rate of about 10%. However, YiKong has a lot of cash in its account, so it can pay the full amount upfront without any interest costs and even receive a 5% discount. All in all, YiKong’s equipment acquisition costs are about 10% lower than those of traditional engineering companies.
With the cash flow and profit margin from its traditional business, YiKong can now cooperate with banks to gradually build up its credit. YiKong has already applied for a loan amount from a top commercial bank and will consider receiving investment from the bank. As YiKong’s valuation gradually rises in the capital market, banks will enjoy substantial returns from their investments, making them more willing to issue loans.
As self-driving technology becomes more mature, “human driving” is also essential.
Several key members of the YiKong team come from mining enterprises and mining research institutes and are very familiar with mining scenes. They all agree that the mining environment for autonomous driving needs to be more standardized than traditional transportation; otherwise, the technology cannot be implemented on a large scale.
This means that the scene needs to be structurally transformed first, which will be completed by engineering companies.
However, no matter how the scene is transformed, there will always be extreme working conditions that self-driving technology cannot overcome, or perhaps self-driving algorithms can solve them after reaching a certain level of advancement and being paired with better sensors. But economically speaking, it may not be worth it.
“For some scenes, it is difficult even for traditional manual driving, so why insist on autonomous driving to solve this problem?” said Sun Qingshan, chief engineer. If a lot of time is spent solving these extreme working conditions, the overall landing progress will be slow.
YiKong’s algorithm director said, “We don’t intend to use autonomous driving to solve 100% of the problems. Waymo has assigned engineers specifically to deal with squirrel recognition. We can avoid these problems in closed scenes and not waste time on algorithms that are largely unnecessary.”According to the algorithm director, their basic idea is to gradually explore how to standardize the scene in the process of traditional business and explore a set of production and construction standards suitable for unmanned driving. Then, let unmanned driving handle the scenes that are suitable for structuring, while the traditional transportation team should handle the Corner Cases that cannot be structured.
Hao, who provides scene-related knowledge input for the easy-to-control R&D team and comes from a mining company, also believes that “it is difficult for the mining scene to be 100% unmanned”.
Watson also holds the same view. Watson also mentioned that doing unmanned driving transportation and traditional “man-driven” transportation on the same mine is conducive to comparing various data, finding the space and boundary for the advancement of unmanned driving technology, and the goal is clearer.
From the perspective of mining companies, this idea is quite wise. After all, what mining companies care most about is whether you can operate safely and efficiently, not whether it is “manned” or “unmanned”.
In fact, Easy-to-Control Intelligent Driving is not the first company to do this and will not be the last one. Uber and Didi have long proposed to “mix dispatch” Robotaxis and traditional ride-hailing cars in the future, and assign orders for simple scenes to unmanned cars and other scenes to manually-driven vehicles. Alibaba, Jingdong, and Meituan’s unmanned vehicles, in the commercialization stage, will also do the same.
Different from Didi, Uber, Alibaba, Meituan, and Jingdong which rely on their own scenes, Easy-to-Control Intelligent Driving does traditional transportation, which means it creates its own scenes.
We may also make such a conjecture: someday, easy-to-control may subcontract the transportation business of a single section of a single mine or a small mine, do unmanned transportation where it is suitable for unmanned driving, and traditional transportation where it is unsuitable, and seamlessl connect them.
At this time, those companies that cannot solve the problems that cannot be solved by unmanned driving and have not laid out “man-driven” businesses in advance will feel “great pressure”.
Why not use large mining trucks?
Whether it is traditional transportation business or unmanned driving business, Easy-to-Control Intelligent Driving only uses wide-body vehicles (usually with a load capacity of about 40-60 tons and a price between 600,000 and 1 million), instead of large mining trucks (usually with a load capacity of 200-300 tons and a price between 20-30 million). This is the choice made by Easy-to-Control after conducting in-depth research on the current situation of mining operations.
Theoretically, large mining trucks are more in line with the national trend of “equipment specialization” for mining, and the operating efficiency is higher. However, in practice, many factors make large mining trucks not the ideal “carrier” for unmanned driving, because the market for large mining trucks is shrinking due to the following reasons:
(1) The cost of purchasing is too high.The purchase cost of a large mine truck is as high as 200-300 million yuan. Normally, every 3-4 large mine trucks correspond to one electric shovel, and the cost of an electric shovel is usually around 100 million yuan. If a mine buys 10 sets of equipment, the total cost will be close to 2 billion yuan, while the recovery period is more than 8 years.
Although the comprehensive operating cost of large equipment will decrease over time, such high prices can only be afforded by large state-owned enterprises. How can cash-strapped private enterprises bear it?
(2) Low delivery efficiency
Wide-body trucks can be loaded onto flatbed trailers and transported to the mining area as a whole, which can be put into use quickly; however, large mine trucks can only be delivered in parts, and the manufacturer dispatches engineers to assemble the parts. A few years ago, a mine in northeastern China bought 8 Trex mine trucks, and it took many workers half a year to assemble them. How can mines wait in times of rising coal prices and urgent production tasks with such efficiency?
(3) High operating costs
Although the work efficiency of large mine trucks is higher and their service life is 3-4 times that of wide-body trucks, they also have higher daily operating costs. In the same area, using wide-body trucks and small excavators to strip each unit of soil costs 9 yuan, but using large mine trucks and electric shovels to strip each unit of soil costs 15 yuan.
When coal prices are high, using large mine trucks is still viable, but when coal prices fall, the cost performance disadvantage of large mine trucks becomes apparent. Some mines found it unprofitable to use large mine trucks and large excavators after calculating the cost, so they switched to using wide-body trucks. “I can’t lose money.” Therefore, large mine trucks have to be idle.
(4) High maintenance costs
I have seen many large idle mine trucks on the scene of an open-pit mine, and their tires look particularly worn out. According to a person who has worked in a traditional engineering company for many years, if a mine truck is idle for a period of time, the tires will age and have to be replaced. A new tire costs at least 180,000 yuan, and according to the calculation of 6 tires per vehicle, it costs 1.08 million yuan.
Apart from the cost, the process of replacing tires is also very complicated. Unlike changing tires for ordinary cars, jacking up the truck is not enough. The tires of a large mine truck weigh 1 ton, and specialized equipment is needed to assist in replacement. The red device in the following figure is the tool used to assist in changing the tires.
In addition to changing tires, large mine trucks have many other maintenance needs.This individual mentioned that over a decade ago, the Caterpillar mining trucks they used required major repairs every three to four years, costing 2 million yuan per repair, with yearly maintenance expenses of 1.5 million yuan. “It’s definitely more expensive now.”
5. Longer maintenance period
Generally, wide-body vehicles are easier to maintain, while large mining trucks not only have high repair costs, but also have longer maintenance periods – sometimes requiring manufacturers to send engineers from abroad to repair, which could take months. This greatly affects the equipment’s attendance rate, which in turn has a significant impact on the continuity of mining operations.
While other unmanned driving companies act as technical solution providers and can still cooperate with holders of large mining trucks, the heavy asset-based model of EHang Drive makes it impossible to deal with the above-mentioned issues relating to large mining trucks.
In fact, given these drawbacks of large mining trucks, since the introduction of wide-body vehicles in 2005, more and more mining companies and engineering corporations have gradually converted to them.
Of course, for the following reasons, many companies still use large mining trucks: A) The safety of large mining trucks is higher, and policy mandates that large state-owned enterprises must use large mining trucks; B) When using large mining trucks, mining companies require far fewer drivers to manage, resulting in significantly lower personnel management costs; C) The entire operational process of the mining enterprise is designed to the standard of large mining trucks, making it difficult to change easily.
However, with the safety of unmanned driving technology surpassing that of human drivers and being able to completely eliminate the need for security staff, factors A and B will become ineffective, and for newly developed mining areas, factor C will not apply. This means that in the future, the few advantages of large mining trucks will also be eliminated by unmanned wide-body vehicles.
A person employed by a mining enterprise said: Although from a policy perspective, equipment enlargement is a trend, from an economic perspective, in the next 10-20 years, wide-body vehicles will still dominate.
From the perspective of EHang Drive, if the market for large mining trucks is shrinking, why would they base unmanned driving technology on large mining trucks?
Mining truck unmanned driving solutions are difficult to standardize and have high marginal costs
Due to the aforementioned reasons, the incremental market for large mining trucks in China is limited, and some companies developing unmanned driving solutions for large mining trucks focus on the stock market.
None of these large mining trucks have wire-controlled chassis, making retrofitting difficult. Furthermore, due to different levels of wear and tear from prolonged use, the mechanical parameters of each old vehicle may be different, so modification can only make a particular truck reach an optimal state, but the parameters of the truck that is modified cannot be replicated in other vehicles.
The consequence of mining truck retrofitting solutions being non-reusable is that: 1. It consumes a lot of manpower and affects efficiency; 2. The marginal cost of the project is particularly high.
In contrast, EHang Drive uses same-power wide-body vehicles with self-contained wire-controlled chassis, making it easy to mass produce and replicate modifications as they are all new vehicles with consistent mechanical parameters.
Why not develop Domain Controllers in-house?## Introduction
Unlike most self-driving solution companies, EasyMile had previously used industrial computers in place of domain controllers until the arrival of their Technology Vice President, Qiao Lin, in April 2020, who initiated a collaboration project with Huawei.
People familiar with Huawei’s car business unit can easily note that Huawei does not sell standalone autonomous driving chips but instead sells integrated domain controllers. This means that by working with Huawei, EasyMile Smart Mobility will no longer need to devote resources to developing its own domain controllers.
Given that most self-driving companies emphasize the importance of “autonomous controllability,” EasyMile’s decision to abandon self-developed domain controllers seems to be unique and not everyone can understand the rationale behind it. Countering outsiders’ confusion, Watson, in an interview with Jiuzhang Intelligent Driving, explained his thought process:
“For mining unmanned driving companies, domain controllers are a relatively common integrated technology that does not have to be self-made. Additionally, this track is small, and whether self-developing domain controllers are worth it is also a question.”
Personally, I agree with this explanation. Furthermore, in my opinion, even if they had designed their own domain controller, EasyMile Smart Mobility, a small start-up company, may not have the ability to overcome a series of complex engineering problems in bulk integration. Thus, in the end, they may have to collaborate with a company that has strong engineering capabilities. Watson also admitted to this point.
Huawei’s MDC President, Li Zhenya, said: “The autonomous driving computing platform is not just about getting the functionality right. The further down we go, the higher the requirements for platform stability and reliability become. Stability and reliability require long-term and unremitting accumulation.”
Qiao Lin also stated: “High-speed and high-quality iteration is key to the success of autonomous driving, and a good computing platform is the absolute foundation for high-speed and high-quality iteration. Without a solid foundation, even the best algorithm is just a castle in the air. The functional safety and information security of the computing platform are systematic issues that require in-depth interactions between algorithm companies and computing platform companies to achieve logical verification between each other.”
Therefore, choosing to use Huawei’s MDC platform is a very practical solution.
For Huawei, supplying the MDC platform to EasyMile Smart Mobility is not just a simple turnkey solution, but also a collaborative process to improve products based on L4 scene requirements. During the cooperation process, developers from both sides think from a product landing perspective. EasyMile Smart Mobility’s development team not only uses the product but also proposes computing platform system architecture requirements based on research and testing progress. In response, Huawei’s development team will integrate these requirements with existing platform capabilities to iterate product development.
Despite EasyMile Smart Mobility’s relatively small size, Huawei’s MDC department considers them to be a benchmark customer in the commercial vehicle industry, providing significant support. Over the past year and a half, Huawei’s MDC department has hired more than 20 people to support EasyMile Smart Mobility.The cooperation between the two parties has achieved very good results in the progress of technology and products through complementing each other’s advantages. Currently, all vehicles of Yeykon Intelligent Driving are operating in standardization testing with MDC.
For Yeykon Intelligent Driving, working with Huawei is better than spending a lot of time on self-developed domain controllers. They can save resources and put them into polishing algorithms (including scheduling algorithms), thus accelerating the landing speed of technology.
Starting from the simplest scenario, whether it is not doing large mining trucks or self-developed domain controllers, Yeykon Intelligent driving consciously stays away from the “high-end” and practically does its most important things well. Similarly, Yeykon’s experimental operation of the unmanned vehicle fleet also avoided complex scenarios and started from the simplest scenario.
Before June 2020, the four unmanned vehicles put into operation by Yeykon at the Hanggaigou coal mine in Ordos were only for testing and could not generate revenue. Since then, Yeykon has moved its unmanned vehicle fleet to a northwest open-pit mine, and the total number of vehicles has gradually increased to 12. Here, the mine pays Yeykon transportation fees, which is the first stop for Yeykon’s commercialization of unmanned driving.
Hanggaigou’s coal mine scenario is relatively complex, the most typical being the many ditches, and there is a long stretch of one-way road. In contrast, the scenario in the current trial operation location of the northwest mine is much simpler-although it is a “mountain,” it looks like an “endless plain,” with extremely wide roads.
On the day of arriving on site, the first question I asked Watson was: testing in complex scenarios is beneficial for obtaining high-quality data quickly, and can accelerate algorithmic progress, which has become an industry consensus. But now you have put all your resources into a simple mining area, which is not conducive to algorithm training. In the long run, will this be a wrong approach?
Watson calmly replied: From a technical point of view, your understanding is correct, but if you look at it from a business perspective and a global perspective, it is different-unmanned driving’s large-scale operation will not start in complex scenarios, but from “friendly” simple scenarios. In fact, the “simple scenario” we are currently testing is the real scenario we will face in the future when commercializing.
Moreover, considering the responses of Chairman Zhang Lei and Sun Qingshan, the general engineer of the mine, etc., in complex scenarios, although it is easy to obtain high-quality data, for a considerable period of time, the mine will only give you a small area to test, and you cannot operate on a large scale. However, in the “friendly” simple scenario, technology can support large-scale trial operation. From the perspective of an unmanned driving company, the difference between testing and trial operation is still very important.- Testing is a way to refine skills through free labor for others, without generating revenue. Moreover, if the scenarios are too complex, there can be significant uncertainty about when revenue can be generated and commercialization may be distant. This is a key reason for talent loss in many autonomous driving companies including Waymo. Trial operation can generate revenue even if the scenario is simple. Moreover, a simple scenario is conducive to quickly expanding the scale of the fleet, which lets research and development personnel see that their work can generate economic benefits quickly, thus mobilizing their enthusiasm.
- Speaking of Waymo’s recent expansion into more complex scenarios in San Francisco but not increasing vehicle deployment in the Chandler area in Phoenix where the scenarios are simple, Watson said his understanding is that Chandler is a sparsely populated remote suburb, where most families have cars, and most stores, offices, and homes have ample parking. Even if Waymo increases its vehicle deployment there, it cannot make money. On the other hand, the situation is different for us. We currently have two cooperated mines with annual production capacity of 20 million tons and 30 million tons respectively, and the potential for further expansion is large. We rank among the TOP 5 open-pit mines in the country. Every time we add a set of equipment, revenue increases.
- I agree with Watson’s idea. Emphasizing the value of complex scenarios for technological progress is difficult to verify. Moreover, even if the technology advances, talent may be lost due to fatigue, which may not be worth the cost. Perhaps, “making money first” is more important. In addition, the idea that “testing in simple scenarios leads to slower technological progress and, as a result, slower commercialization” is a cognitive misconception. Many domestic autonomous driving companies often emphasize that “the road conditions in China are more complex, so training algorithms in China will advance faster, meaning that the commercialization of autonomous driving technology in China can surpass that of the United States.” This logic is flawed, as the “practice problems” you do are more complicated than others, but your “formal exam questions” are much more complicated. Conversely, American companies have simpler “practice problems” and their algorithmic progress may be slower, but their “formal exam questions” are also easier.
- Now, Yikong Zhi Jia has chosen to conduct trial operations in large open-pit mines with simple scenarios, which is a typical case of “simple practice problems but also simple formal exam questions.” Can you believe that their “algorithmic progress is slow and not conducive to commercialization?”
- Currently, Yikong has deployed two sets of equipment (each set consists of one excavator and six wide-body vehicles), and the third set of equipment will be in place soon. Watson said: “With the increase in fleet size, we can quickly discover more problems and then tackling more complex scenarios will be easier.”Below is the translated Markdown text in English, with HTML tags preserved for professional purposes. Only corrections and improvements have been made without further explanations:
The statement “being able to quickly discover more problems” has also been indirectly validated by the supplier. On the day of my visit, both the vehicle supplier Shaanxi Tongli and the LiDAR manufacturer Ouster had engineers present at the scene. These two companies have partnerships with many unmanned mining companies, but they receive the most feedback on line-controlled chassis and LiDAR issues from E-Control’s intelligent driving system.
On the same track, the company that can help suppliers “identify” more bugs must be the one with the strongest testing intensity, and its technological progress will not be slow.
- Evaluating the ability of an unmanned mining company is not only about its level of intelligence relative to that of a single vehicle, but whether it can truly integrate into the entire mining operation system like “having a driver.” However, if it only does testing, the unmanned driving company will find it difficult to truly become a part of the production process in the mining area. In contrast, during trial operation, unmanned driving becomes a part of the entire mining operation system and is easier to obtain more resource support.
From E-Control’s practical experience, during the testing phase, the mining area does not care about your testing efficiency and technological progress. For example, they only allow you to test in a small area and provide you with a not-so-good excavation machine. If they need it urgently halfway through, they will take the machine away, and at this point, your unmanned driving test will have to “take a break.” But during the trial operation phase, because unmanned driving becomes a part of the entire mining operation, they value it highly and are willing to sacrifice economic benefits to support the achievement of the task.
Currently, in the two mining areas where E-Control collaborates, the mining areas provide great support to E-Control—not only by providing free offices and warehouses, but also by accepting that the operational efficiency of unmanned driving is currently inferior to that of manual labor. For the mining area, this means a considerable loss of profit.
E-Control Chairman Zhang Lei said that in the B2B business, benchmarks are very important. The mines they cooperate with now have a great influence in the industry. The unmanned vehicle team has carried out large-scale operations in these two mines, and it will be easier to expand to more complex scenarios of small and medium-sized coal mines. “The technology is easy to difficult, but management is difficult to easy.”
Zhang Lei believes that as the scenario expands and the scale of the vehicle team expands rapidly, the barriers of unmanned driving algorithms (single-vehicle intelligence) will become lower and lower.
However, an unmanned driving team in mining operations relies not only on single-vehicle intelligence but also on collective intelligence (scheduling).
A chief engineer from a certain open-pit mine production technology department said: “The biggest difficulty in mining operations lies in multi-grouping and path planning (in fact, it refers to scheduling). Existing scheduling software is not practical enough, and the variables are not enough to meet the requirements of operations.”Usually, a big working face has 6 or 7 teams of overlapping operations. This means that there could be dozens of cars intersecting at some intersections, and how to schedule them safely and efficiently is a key issue. For example, the vehicles cannot be too close together, but the loading efficiency must also be ensured.
Many team members from mines, including Sun Qingshan, the chief engineer, all agree that unmanned driving surpasses human driving in terms of operational efficiency, not relying on the intelligence of a single vehicle, but on a mixed scheduling algorithm. That means that in the long term, scheduling algorithms are more valuable for improving production and management efficiency than unmanned driving vehicle intelligence algorithms.
As mentioned earlier, Brother Hao said that under traditional operations, the shovel-to-truck ratio is fixed, for example, one excavator corresponds to six trucks at a certain hauling distance. With scheduling algorithms, if there are two groups of overlapping operations, each group will not have a fixed number of six trucks, but two excavators and twelve trucks. This kind of mixed scheduling is clearly more efficient.
Next, Yi Kong’s goal is to go for safety and double-group operations. On the surface, they have chosen a “condition-friendly” simple scenario, but after going for safety and double-group operations, the actual challenges are still significant.
Sun Qingshan said, “Scheduling includes the coordination between cars and cars, as well as cars and excavators, which is the most important part of mining transportation. However, many unmanned driving companies have not reached this stage. Everyone’s focus is still on single-vehicle intelligence.”
Yi Kong’s early efforts in improving scheduling algorithms have shown preliminary results.
According to an engineer from Yi Kong, the position of the excavator is always changing in the soil-loading area. Logically, the position of the vehicles should also be adjusted accordingly to improve the loading efficiency. However, the more common practice in the industry is to “keep the vehicle’s parking position unchanged as long as the map does not change.” In contrast, at Yi Kong, the vehicles always park at the closest point to the excavator, thus improving the efficiency of the loading process.
In addition, the engineer also stated that whether the path planning is done based on a single vehicle or a comprehensive scheduling system will also affect the efficiency. “In the industry, only one vehicle can enter the soil-loading area at a time, while at Yi Kong, four vehicles can enter the area simultaneously.”
This is actually due to the difference in scheduling algorithms.
Yang, the on-site responsible person for a certain open-pit mine contractor, said that when they first cooperated with Yi Kong, the speed of soil loading and unloading was relatively slow. After optimizing the scheduling algorithm, the efficiency has improved significantly. Currently, under the condition of continuous 24-hour operation, unmanned driving efficiency has been greatly increased.
Of course, the engineers at the Yi Kong site believe that due to the high level of safety redundancy, there is still room for improvement in the path planning for the soil unloading area, and there may occasionally be more complex paths.
Unlike training perception and decision-making algorithms, where “the more complex the scenes, the better”, training scheduling algorithms is “the larger the fleet size, the better”. This means that if Yi Kong focuses on complex scenarios during trial operations, the scheduling algorithm will not receive effective training since the fleet size cannot be increased. However, rapidly expanding the fleet size on a simple mining site in the northwest is just conducive to the iterative optimization of the scheduling algorithm.In addition, Lin Qiao, Vice President of Easy Control Technology, pointed out another benefit of quickly scaling up the fleet in simple scenarios: to ensure the attendance rate of operation vehicles, spare parts are stocked and ready for repair and replacement at any time, and the more vehicles there are, the lower the required spare parts ratio.
When discussing the topic of “whether starting from simple scenarios is unfavorable to technological progress” with the Easy Control Intelligent Driving team, the author’s deepest feeling is that Zhang Lei and others do not stick to one or two points, but think about “how to make the company’s commercial ability more competitive” with a global perspective.
In the recent popular article “Four Dark Logics of China’s Internet”, the author Wei Xi mentioned such an example:
In 2015 and 2016, after the short video trend took off, the leading short video platform Meipai became the darling of the investment community, and its parent company Yixia Technology received five rounds of financing, with the last series E financing reaching a scale of 500 million US dollars.
However, after the rise of Douyin and Kuaishou, Meipai and Meipai, which were once so impressive, were quickly marginalized.
Why did Meipai and Meipai lose to Douyin? The author’s explanation is that their focus is different from Douyin:
When Meipai was born and developed, China’s 4G network was not particularly mature, so the technical requirements for how to play in different network environments were very high, and in the cognitive framework of Meipai’s founder Han Kun, technology was an important barrier, so he naturally paid less attention to other variables that truly decided the core of this race.
However, by 2017, with the maturity and decline of 4G tariffs, technical barriers were proven to be vulnerable, and recommendation algorithms and content production ecology became the core competitiveness of short videos. In a sense, perhaps the outcome of this war had already been doomed from the beginning.
Returning to the autonomous driving path in the mining industry, many entrepreneurs and investment institutions still believe that technology is the most important barrier. However, when it comes to “rapid scale-up,” they may find that “technology is important, but it is not enough to have technology alone.” A vice president of a head investment institution also mentioned in a recent communication with Easy Control Intelligent Driving that he does not consider technology to be a barrier and values the team’s operational and market development capabilities.
Moreover, as we analyzed earlier, in the long run, Easy Control Intelligent Driving’s choice of starting from simple scenarios, which seems unfavorable to technological progress, will ultimately be more conducive to technological progress.
Will they be “downplayed” by the autonomous driving companies in the “big field”?
Like autonomous driving companies focused on end logistics and ports, companies focused on autonomous driving in mines are often asked by outsiders: “Once the market matures, will you be downplayed by companies engaged in Robotaxi or trunk logistics?”
For those who are really involved, the answer is very clear: no.On one hand, for companies that focus on Robotaxi or long-haul logistics, doing business in the “small scene” of mining is just too costly in terms of opportunity cost – it not only affects their main business, but also the company’s valuation. Some Robotaxi companies are worried that L2 will affect their valuation, let alone mining autonomous driving, which is at the “lowest end of the totem pole”.
On the other hand, even if the so-called “better companies” are really interested in entering such “small scenes” like mining, the team’s understanding of the scene is the core competitiveness in technology implementation, and understanding a particular scene well requires even the most capable people to spend enough time.
Moreover, many of the technical backbone of Robotaxi companies are simply unwilling to go to the mines – in previous recruitment, Easymile has communicated with many engineers from Robotaxi companies and found that 70% to 80% of them are unwilling to go to the mines; even some who initially “agreed to go” changed their minds when the offer was approaching.
Since the technical backbone of Robotaxi companies is unwilling to go to the mines, their understanding of the scene is insufficient, and the “dimension reduction blow” is impossible to carry out.
Let’s talk about a few “aside” paragraphs first:
- Before paying attention to the autonomous driving industry, the author was not very familiar with the term “scene”. At that time, the author often wondered: the various medical system apps look very powerful, is the reason why the app from Hospital A is better than the app from Hospital B because the programmers they hired have stronger software capabilities? No, the truth is that the programmers they hired are more familiar with the business system of the hospital.
Later, after knowing the term “scene”, the author realized that the statement above can be simplified as “Whether the programmer can write a good app or not, software ability is only one aspect, his understanding of the scene may be more crucial.” A lot of code in Alibaba Cloud is also written in the customer’s workshop, and that’s the reason.## 2. In August 2017, during a period of unemployment before entering the autonomous driving industry, I once intended to join some industries that are now considered very mediocre. However, several interviews made me particularly frustrated. I found that some “interviewers” were very picky about me, and their vision, thinking ability, basic cognitive ability, and language skills were completely different from mine. However, if I got in, these people would be my “superiors”.
Later, I felt that the biggest embarrassment of mid-career switchers is that people with comprehensive qualities that may be far inferior to yours will become your leaders. And the reason why they can become your leaders is not because they are better than you, but just because they have stayed in this industry longer than you. This is a typical case of “defeating people” through understanding of “scenes” and defeating “higher-quality people”.
3. An example that I often share with my friends is that although my dad did not finish primary school and often reads words incorrectly, I graduated from a 985 university and my comprehensive ability is indeed much stronger than my dad’s. But if I want to run an apple orchard across industries, I have to spend at least five years to reach his current level because he has been doing it for twenty years. So, if you don’t spend enough time understanding the scene, then in specific business, you may become a “high-quality, low-ability” person.
Back to the point. In terms of talent, companies with “big scenes” such as Robotaxi and unmanned trunk logistics do have more talented people, but no matter how talented a person is, if he wants to understand how to write algorithms for small scenes, there is no shortcut, and he still needs to spend a lot of time to endure.
Recently, in an interview with “Cyber Automotive”, Zhu Lei, CEO of White Rhino answered the question of “dimensionality reduction” as follows:
On the one hand, the data collected in different scenes is different, and the complexity of the problems encountered is also different, which will lead to the combination of the entire technology and modules being different;
On the other hand, the interaction problems that need to be solved during different scene operations are completely different. Taking unmanned sweeping vehicles as an example, not only the automatic driving technology such as algorithms needs to change, but the entire team also needs to research how to clean the floor.
In addition, in an interview before HDAuto, high-level executives from more than one Robotaxi company stated that the key to competitive segmentation scenes is actually resource barriers and first-mover advantages. “Once a certain company’s autonomous driving technology is widely used in areas such as parks, airports, and ports, the cost of replacement is very high. Moreover, once a company has mature experience, in a situation where there is little difference in technology and cost, it is difficult for Robotaxi companies to outperform it.”
In fact, very few people in the industry mention “dimensionality reduction” because they really understand it. Those who shout “dimensionality reduction” are basically investors and the media who have not practiced themselves. Especially, the media uses “dimensionality reduction” the most, and truly “know the difficulty without passing through it.”## “Our Barrier Is Our Experience”
As one of the leading companies in the Robotaxi field, Pony.ai did not use the strategy of “dimensionality reduction” in its truck business. Instead, it emphasized that “part of the data can be reused.” Therefore, when expanding into new scenarios, the focus is not whether the technology can reduce dimensionality, but whether the data can be reused. Can the data accumulated in other scenarios be reused in mining scenarios?
A partner of a top investment institution once worried that unmanned mining companies would be hit by companies specializing in trunk logistics, but after visiting mining sites, he came to the conclusion that “there is no such thing as a reduction in dimensionality.”
“Our Barrier Is Our Experience”
As mentioned in the previous paragraph, other companies in different fields cannot execute the strategy of dimensionality reduction in mining scenarios due to a lack of understanding of the industry. For Pony.ai, the key to building a barrier to entry is the spirit of “going up to the mountain and down to the countryside” of its technical personnel.
In October 2020, when writing the history of Alibaba Rhino Smart Manufacturing, Qianhei Technology mentioned a story of “going up to the mountain and down to the countryside” –
In Rhino Smart Manufacturing, all algorithm engineers need to work as team leaders on the production line and personally lead production with the workers.
A Ph.D. in algorithm from Nanyang Technological University also started as a team leader and lived with the workers for two months. His colleagues were shocked: “This is definitely the most expensive team leader in the world.”
The culture of Rhino Smart Manufacturing is that anyone who cannot pass the test of “going up to the mountain and down to the countryside” and cannot learn the “Alibaba flavor” will be seriously criticized or may be dismissed directly.
In the self-driving industry, there are many examples of companies losing their competitiveness due to managers not having this kind of field experience, and the most typical one is Uber ATG. After burning nearly $3 billion, Uber ATG, with its luxurious team, could not achieve anything and had to “sell itself” to Aurora.
According to the technology media The Information, in September 2020, a senior manager of Uber ATG wrote a letter to Uber CEO Dara Khosrowshahi complaining about the many flaws of Uber ATG before leaving the company. One of the main points was that although Uber ATG had a fantastic team, its main intellectual resources were from Carnegie Mellon University, and these people did not have enough field experience.### Translation
Of course, there are also positive cases.
Waymo chose to collaborate with truck industry veterans with more than 20 years of experience in order to understand the scene and integrate their knowledge into their algorithm.
TuSimple is one of the leading companies in the unmanned truck raceway, and TuSimple’s CEO Chen Mo has ridden behind a long-distance truck several times in order to understand the scene.
Last month, when interviewed by “Wandian”, Chen Mo was asked a question: “Some truck counterparts in China say that some roads in China have two drivers driving, and they can remove one driver through L3.” Chen Mo’s answer was: “This statement is particularly absurd. Why two drivers? Generally, they are couples or siblings, because one person is sleeping. The two drivers take turns driving and sleeping. If you remove one driver, will the remaining driver never sleep, or will he sleep without incident?”
When the “Wandian” reporter mentioned that “their counterparts said that L3 can reduce the workload of drivers, and a driver who originally drove for 4 hours needs to switch, they may allow this driver to drive for 8 hours”, Chen Mo’s answer was: “All drivers switch every 4 hours, and the two drivers take turns driving and sleeping. I don’t know if they have driven a car, but I have driven with them.”
A few simple questions and answers easily show that the “CEO who has driven with them” has a much better understanding of the scene than those “CEOs who have never driven with them”.
At ECOVACS, Zhang Lei, co-founder and chairman, spent 80%-90% of his time in the mine; Lin Qiao, vice president of technology, and the director of the algorithm also spent 50% of their time in the mine. Technical personnel based in Beijing headquarters will be dispatched to the mine in batches for support. Sometimes, the Beijing office is almost empty, with “only a few functional departments’ personnel staying behind”.
A test engineer who came from another unmanned driving company in the mine said: “Previously, when I was working on the mountain project, I often did not receive technical support from the headquarters. After coming to ECOVACS, I have not encountered such problems.”
In the commercialization stage of unmanned driving, engineering capabilities are higher barriers to entry than technical capabilities. However, many talented people cannot “get it done” on engineering issues. Because the engineering level involves many tedious and uninteresting details that talented or self-proclaimed “geniuses” usually have no patience for.The most important person for engineering is not the most technically talented individual, but the one who is particularly solid, can endure all kinds of boredom, can repeat simple tasks, and has particularly strong execution skills. If many technical experts in a company have these qualities, then the company will have strong competitiveness in commercialization.
According to Watson’s plan, students who will serve as heads of the marketing department in the future will also be stationed in the mine for more than a year to “ensure that they have a very good understanding of the situation on site and that the information conveyed externally must be in line with the facts.”
Watson often says, “We are the autonomous driving company that eats the most dirt. And the dirt we have eaten is our competitive barrier.”
During the process of interaction with the Yi Control team at the mine, the author suddenly thought of their working state, which is similar to that of Huawei when it captured the African market years ago.
Speaking of so many people all “eating dirt” together, Watson added:
Usually, in a growing company, early members of the team hold a lot of resources, and there are many outside temptations. If there is no strong “revolutionary emotion” within the company, it is easy to have problems such as key members being poached by others for high salaries.
But now we are all working in the mine together, and we have established deep revolutionary emotions with each other. As a result, the team will be much more stable in the future.
At this point, the author mumbled to herself, “Having carried guns together, having been to the countryside together….” But before she could finish, Watson added, “Having eaten dirt together.”
At the time, a colleague present added: One advantage of doing research and development and operation in a remote and closed environment such as a mine is that a group of people can be “trapped” together for several months with no other social activities. Just like military training during university, such an environment is indeed very conducive to cultivating revolutionary emotions.
It is worth mentioning that the colleagues in HR, finance and other functional departments who have experienced living in the mine will increase their empathy and will not easily treat colleagues who have been stationed at the mine for a long time in a harsh manner for the sake of their own small department’s KPI.
It must be said that compared with common company-funded tourism and wining and dining, “eating dirt together” is the best team-building activity for Yi Control.
Is the CEO’s “non-technical background” a “weakness”?
Previously, an investment institution questioned Yi Control’s CEO Watson, who came from a gaming background, stating, “Most founders and CEOs of other autonomous driving start-ups have backgrounds in autonomous driving from big companies. Is it reliable that your CEO Watson is a gaming background?”In Watson’s view, this logic is completely invalid. Who says that someone coming from a gaming background cannot enter the field of autonomous driving? “Tusimple and IMa, the benchmark companies in the field of unmanned logistics, both have CEOs, Chen Mo and Liu Wanqian, who come from a gaming background.”
Some people also question whether your CEO, who has no technical background, can lead an autonomous driving company, unlike other autonomous driving companies whose CEOs come from a technical background.
In the author’s opinion, investors who have such doubts probably have no entrepreneurial experience and do not understand the role of a CEO – they seem to misunderstand what the CEO’s main responsibilities are.
Let’s take a look at these questions:
- Were Liu Bang and Song Jiang highly skilled in martial arts?
- Was Tang Sanzang highly capable of confounding demons and vanquishing evil?
- Did Mao Zedong often lead his troops into battle? (According to an article published by the Military Science Academy on Phoenix Network’s History Channel in November 2014, on the first day of the Lunar New Year in 1929, Mao Zedong laid a trap at Dabaodi and had a life-and-death fight with the National Revolutionary Army’s Liu Shi-yi, “this was the only time that Mao Zedong had ever led a charge with a gun in his life.”)
In a technology company, the CEO’s main role is not to personally create technical solutions but to act as a “glue” that brings together a group of technology experts to develop products. In addition, like Tang Sanzang, the CEO must be focused on the company’s goals and be able to resist various temptations from the outside world.
Currently, Yi-Controlled Autonomous Driving employs five full-time PhDs (some with at least two years of work experience prior to joining the company, including Lin Qiao, the Vice President of Technology who has entrepreneurial experience and previously served as the hardware director of Alibaba’s unmanned driving team), which is more than the total number of full-time PhDs employed by other leading unmanned mining companies.
Those who question the value of a “CEO without a technical background” should consider this question: for a CEO of a mining autonomous driving company, which ability is more valuable, writing code and creating technical solutions or mobilizing a group of PhDs who work long hours in the mines?
In fact, if these doubters had witnessed “expert-type CEOs” with technical backgrounds who were overly confident in their professional abilities, and subsequently forced their subordinates to follow their “technical roadmap” to the letter, leading to frequent “technical disputes” (essentially a battle of egos), and ultimately driving out highly skilled technical staff, they might revise their bias against “non-technical CEOs”.
When asked some technical questions, Watson is particularly frank, saying, “I don’t have much judgment when it comes to technology, so I can talk to Lin Qiao.” Sometimes, he even asks me to talk to a test engineer, stating, “he’s more professional.” When faced with technical problems, Watson consults extensively with technology experts in the industry and then discusses the issue with team members to reach a consensus, rather than forcing others to accept his own “technical roadmap,” while also ensuring that the choice of technical direction does not cause major problems.In addition, as a person who finds it difficult to fall asleep at regular times while on business trips, I have always been curious about how frequent business travelers manage to get rest. As it turns out, a friend from ECOVACS Robotics team told me, “Watson sleeps very well. He can fall asleep on the plane whenever he wants, and even if he drinks coffee and tea at night, he can still fall asleep before 11 o’clock.”
What’s more enviable is that Watson frequently shares a room with a colleague who snores loudly, but his own sleep is not affected in the slightest.
Usually, only people with a good mentality and no anxiety can have good sleep.
Compared to many friends around me, I am already one of the least anxious and most mentally stable, but when it comes to people like Watson who can “fall asleep before 11 o’clock even after drinking tea,” I can only admit that there are always higher mountains.
It is said that decision making can be blind when people are anxious.
For example, a leading international car company’s “buy, buy, buy” in the autonomous driving race has never stopped, and they have spent over $6 billion in the past five years on investments, many of which have been duplicated or of questionable quality. The key reason is that the chairman, a third-generation family head, is particularly anxious and worried about “wealth not lasting for more than three generations” happening to him.
I believe that this leading company’s “buy, buy, buy” in the autonomous driving race is a kind of “anxiety tax,” and “there is no essential difference between getting logic thinking courses from an anxious young person and being part of this company.” Sure enough, not long ago, the company announced deep cooperation with Mobileye in the ADAS business, admitting that many of their past investments were not successful.
In contrast, people who are not controlled by anxiety are more likely to stick to long-term values when making decisions.
Some anxious entrepreneurs like to do some “project for the VCs,” painting a rosy picture for investors and customers, but Watson’s approach is to allow investors to freely board the car when they come to inspect the plant, without any preparation work; and resolutely not tolerate business people to draw cakes for customers, because “the cakes currently drawn are pitfalls buried for the future.”
Previously, some investment institutions were worried that ECOVACS Intelligent Driving would obtain high valuations based on the “concept” of unmanned driving, but now they are doing traditional transportation businesses. If they make a lot of profit from traditional transportation businesses, will they “forget their original intention”? However, in my opinion, if these investment institutions have enough understanding of “long-termism” and Watson’s “good sleep,” they would not have such concerns.
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.