Read More《自动驾驶:万亿赛道的终极博弈，下一个十年谁主沉浮?》
Positive Comments: Technological Iteration and Ecosystem Building, Autonomous Driving is Approaching the Commercialization Threshold
Autonomous driving is known as “one of the most difficult problems in the AI field.” However, judging from the current industry trends, the triple resonance of technological breakthroughs, ecosystem improvement, and policy pilots is propelling it to accelerate from the “laboratory stage” to “large – scale commercial use.” Clear positive signals have emerged in this ultimate competition on the trillion – dollar track.
First of all, the diversified exploration of technological routes has injected innovative vitality into the industry. The debate between the “pure vision school” and the “multi – sensor fusion school” mentioned in the news is essentially a choice between two paths of technological innovation: Tesla’s pure vision route uses “data – driven + end – to – end algorithms” to feed back model iterations with low cost and massive data. FSD V12 has achieved “human – like decision – making” in complex scenarios. On the other hand, the multi – sensor fusion routes of Waymo and Huawei, through the collaboration of lidar, millimeter – wave radar, and cameras, demonstrate higher safety redundancy in scenarios such as extreme weather and low – light conditions. These two routes are not in opposition but complementary. The pure vision school uses algorithms to make up for hardware shortcomings, while the fusion school solves the mass – production problem through cost reduction (the cost of lidar has dropped from hundreds of thousands of yuan to the thousand – yuan level). This “technological competition” pattern has accelerated the expansion of the overall technological boundaries of the industry. As the Gelonghui Research Institute said, “The end of perception hardware is algorithms.” Whether it is the popularization of the BEV + Transformer architecture or the application of large models in decision – making and planning, it confirms that the improvement of algorithm capabilities is becoming the core driving force for technological breakthroughs.
Secondly, the maturity of the industrial chain ecosystem provides support for large – scale implementation. The four major ecological levels of “perception layer – decision – making layer – execution layer – support layer” dissected in the news have formed a complete technological closed – loop. At the perception layer, the performance of lidar from domestic companies such as Hesai and RoboSense is approaching international standards, and companies like Sunny Optical hold an important share in the field of automotive – grade lenses. At the decision – making layer, NVIDIA’s Orin chip (with a computing power of 2000 TOPS) has become the preferred choice for L4 – level solutions, and domestic chips such as Horizon Journey 5 and Black Sesame A1000 have filled the gap in computing power. At the execution layer, companies such as Bosch and Desay SV have achieved mass production of wire – controlled chassis, solving the reliability problem of “from decision to action.” At the support layer, 5G, V2X technologies, and simulation platforms (such as Waymo’s simulation tests covering billions of miles) have significantly reduced the cost of road tests. More importantly, the Chinese industrial chain has achieved a leap from “following” to “running side by side” in multiple links – the cost reduction of lidar, the breakthrough in computing power of domestic chips, and the autonomy of wire – controlled chassis have all given domestic enterprises the initiative in global competition.
Finally, policy pilots and commercial exploration have injected confidence into the industry. Although large – scale commercial use of L4 – level autonomous driving is still restricted by regulations, “small – step and fast – paced” pilots have been launched at home and abroad: Germany allows L3 – level vehicles on the road, Beijing and Shanghai in China have opened up Robotaxi tests, and companies such as Baidu Apollo and WeRide have achieved “fully driverless taxi” operations in specific areas. In the logistics field, autonomous trucks in closed parks and last – mile delivery vehicles have entered normal operation. These pilots not only verify the reliability of the technology but also promote the adaptive adjustment of policies. For example, some domestic cities have issued the “Administrative Measures for Road Testing and Demonstration Applications of Intelligent Connected Vehicles,” clarifying the rules for determining accident liability. The EU has passed the “General Safety Regulations,” incorporating autonomous driving systems into the vehicle safety certification system. The “two – way interaction” between policies and commercialization is opening up broader imagination space for the industry.
Negative Comments: Technological Bottlenecks, Policy Lags, and Ethical Risks, the Industry Still Needs to Overcome Multiple Obstacles
Although autonomous driving shows a strong development momentum, it still faces multiple challenges from “partial breakthroughs” to “full – scale popularization.” If these challenges are not properly addressed, they may delay the industry’s commercialization process and even trigger a periodic trust crisis.
Technological reliability remains the biggest shortcoming. Currently, the implementation of L4 – level autonomous driving is mostly limited to “restricted scenarios” (such as specific cities and fixed routes), and its ability to handle complex dynamic environments (such as heavy rain, nights without streetlights, and sudden crossing pedestrians) is still unstable. The pure vision route relies on the “human vision simulation” of cameras, but it is prone to misjudgment in scenarios such as strong light and foggy days. Although the multi – sensor fusion route improves redundancy, the “point cloud recognition” of lidar may still fail in extreme weather (such as the scattering of laser beams caused by heavy rain). More importantly, the “generalization ability” of algorithms has not been fully broken through. Even after training with massive data, the model may still make decision – making mistakes in “long – tail scenarios” (events with extremely low probability but serious consequences). For example, in 2023, an autonomous driving test vehicle caused a collision accident at an intersection due to misjudging the intention of a pedestrian, exposing the deficiency of the algorithm in dealing with “non – standard behaviors” (such as a pedestrian suddenly turning back). The lack of technological reliability directly affects the public’s trust in autonomous driving and also restricts the expansion of the commercial scale.
The lack of uniformity and the lag of policies and regulations have become a “bottleneck” problem. The popularization of autonomous driving requires cross – regional and cross – national regulatory coordination, but the current standards vary significantly among countries: Germany requires drivers to “take over at any time” for L3 – level vehicles on the road, while California in the United States allows L4 – level fully driverless taxis to operate. In China, different cities have different requirements for Robotaxi test licenses, operating ranges, and data reporting, which forces enterprises to adjust their plans for each city separately, increasing the operating cost. What’s more troublesome is the determination of accident liability. When an autonomous driving vehicle causes an accident, is the responsible party the car manufacturer, the algorithm provider, or the vehicle owner? Currently, only a few countries such as Germany have clearly stated that “car manufacturers are responsible for L3 – level accident liability,” and most regions are still in a “legal vacuum.” The ambiguity of policies not only increases the compliance cost of enterprises but also may cause legal concerns among consumers, hindering the improvement of market acceptance.
There is still no social consensus on ethical and safety risks. The controversy over the “trolley problem” (such as how the algorithm chooses the collision object when an accident cannot be avoided) remains unresolved, and there are significant differences in the public’s acceptance of “machines making life – and – death decisions.” In terms of data privacy, autonomous driving vehicles need to collect sensitive information such as location and camera images in real – time, and if leaked, it may pose risks to personal privacy or national security. In terms of network security, autonomous driving systems rely on software control, and hacker attacks may lead to vehicle hijacking (such as a test vehicle being remotely controlled to turn due to a system vulnerability in 2024). Solving these problems requires the joint promotion of technology (such as encryption algorithms and security chips), law (such as data classification protection), and social consensus, and it is difficult to completely eliminate them in the short term.
In addition, investment risks cannot be ignored. Autonomous driving is a typical “long – cycle, high – investment” track. It takes 5 – 10 years from technology R & D to mass production and implementation, and the annual R & D investment of a single enterprise exceeds 1 billion US dollars (for example, Waymo invests about 3 billion US dollars annually). Moreover, in the initial stage of commercialization (such as Robotaxi), there is a double pressure of “high operating cost + low order volume” (the cost per order may exceed 50 yuan, while the user’s willingness to pay is only 20 – 30 yuan). Although the capital market is optimistic about the industry’s prospects, excessive speculation on “concept stocks” may lead to valuation bubbles. In 2023, the stock price of a lidar concept stock plunged by 60% within three months due to the failure to meet the expected technological progress. For investors, if they blindly chase hot spots while ignoring technological barriers and commercialization paths, they may face huge losses.
Advice for Entrepreneurs: Focus on Scenarios, Deeply Cultivate Technology, Collaborate in the Ecosystem, and Build a Moat in Niche Markets
Although the trillion – dollar track of autonomous driving has broad prospects, for entrepreneurs, the “timing of entry” and “strategic choices” are crucial. Based on the current industry situation, the following advice can be considered:
Focus on niche scenarios and avoid the “big – and – all – encompassing” technological route. The large – scale implementation of L4 – level fully autonomous driving still takes time, but specific scenarios (such as logistics in closed parks, last – mile delivery, and port freight) are already commercially viable. Entrepreneurs can prioritize “high – demand, low – complexity” scenarios. For example, port freight has low speed requirements for vehicles and fixed routes, and autonomous driving can significantly reduce labor costs. Last – mile delivery (such as community express delivery and supermarket fresh produce delivery) has high – frequency demand, and there are relatively loose restrictions on vehicle size and speed. By focusing on niche scenarios, enterprises can quickly verify the reliability of the technology, accumulate operational data, and lay the foundation for subsequent expansion into more complex scenarios.
The choice of technological route needs to balance “cost and performance”. The pure vision route is suitable for enterprises with rich data accumulation and strong algorithm capabilities (such as car manufacturers with a million – level fleet). The multi – sensor fusion route is more suitable for scenarios with high safety requirements (such as Robotaxi and heavy – truck transportation). Entrepreneurs need to choose the technological route according to the requirements of the target scenario (such as weather conditions and road complexity). If the target scenario is mainly sunny days and highways, they can focus on pure vision + low – cost sensors. If it involves complex environments such as rain, fog, and nights, redundant sensors such as lidar need to be added, but at the same time, the overall cost needs to be reduced through algorithm optimization (such as replacing imported lidar with domestic ones).
Deeply participate in ecosystem collaboration to make up for resource shortages. The ecological chain of autonomous driving is long and complex, and it is difficult for a single enterprise to cover all aspects. Entrepreneurs can make up for their shortcomings through “technological cooperation” or “ecosystem alliances.” For example, cooperate with chip manufacturers to customize dedicated chips with low power consumption and high computing power; cooperate with simulation platform enterprises to reduce road – test costs through virtual testing; cooperate with map manufacturers to obtain high – precision map data support. For domestic entrepreneurs, they can focus on the opportunities of “domestic substitution.” In areas such as lidar, automotive – grade chips, and wire – controlled chassis, domestic enterprises already have a technological foundation, and collaborating with the upstream and downstream of the industrial chain can quickly establish competitive advantages.
Plan ahead for policy compliance and ethical risks. Policies and regulations are the “last mile” of autonomous driving commercialization. Entrepreneurs need to study the regulatory requirements of the target market in advance (such as the application conditions for test licenses and data reporting standards) and actively participate in the formulation of industry standards (such as joining the Intelligent Connected Vehicle Industry Alliance). In terms of ethics and safety, they need to actively disclose the decision – making logic of the algorithm (such as through “explainable AI” technology), establish a data privacy protection mechanism (such as using federated learning technology to avoid data leakage), and conduct regular public communication (such as holding open days and publishing safety reports) to enhance social trust.
Carefully plan the financing and commercialization paths. Autonomous driving requires large R & D investments and has a long pay – back period. Entrepreneurs need to reasonably control costs and avoid “burning money for expansion.” In the initial stage, they can quickly obtain income through “technology licensing” (such as providing algorithm solutions to car manufacturers) or “scenario – based operations” (such as cooperating with logistics enterprises to implement autonomous trucks). At the same time, they need to clarify the “key milestones” of the commercialization path (such as reducing the cost per kilometer in a certain scenario to a certain level and achieving a certain annual operating mileage) to convey a clear value signal to investors.
Conclusion: The end – game of autonomous driving may far exceed our current imagination. However, regardless of how the technological route evolves and the ecosystem is reconstructed, “solving real needs” and “creating user value” are always the core. For entrepreneurs, instead of chasing the “ultimate say,” it is better to deeply cultivate technology and build barriers in niche scenarios and find their own space for survival and development in this decade – long competition.
- Startup Commentary”AI Reconstructs the Supply Chain: JD Aims at a Trillion – Dollar AI Ecosystem”
- Startup Commentary”Pre – made Chinese cuisine is surrounding the Japanese dining tables.”
- Startup Commentary”Taobao Flash Sales and In-store Group Buying Are Really Here!”
- Startup Commentary”TikTok Reaches a Turning Point in Its Fate”
- Chain Exploration”In the Web3 wallet war, is MetaMask also confirmed to issue tokens?”
