Technical Route Controversy: Hardware Redundancy vs. Algorithm Intelligence
The development of autonomous driving technology has led to a clear division into two major technical camps: pure vision solutions and multi-sensor fusion solutions. This technical route controversy essentially represents a clash between the development philosophies of “hardware redundancy” and “algorithm intelligence.” The pure vision solution concentrates all pressure on the algorithms, betting on an intelligent future; while the multi-sensor fusion approach focuses more on engineering implementation, opting for a proven practical solution.
Tesla, as a staunch advocate of the pure vision solution, has completely eliminated millimeter-wave radar, relying entirely on cameras and artificial neural networks for autonomous driving. Musk has publicly stated that multi-sensor fusion may lead to “data inconsistency” issues, which could actually reduce driving safety. He argues that if lidar and cameras “disagree,” which one should the system listen to? This itself poses a systemic risk.
Cost-Benefit Analysis: Short-Term Investment vs. Long-Term Returns
From a cost perspective, the pure vision solution has a clear advantage. The cost of cameras is only a few dozen dollars, while early lidar systems cost tens of thousands of dollars. Although prices have decreased, they still require thousands of dollars. Millimeter-wave radar costs only a few hundred yuan, and ultrasonic radar can be as low as a few dozen yuan. This cost difference directly impacts the overall vehicle price and commercialization process.
However, cost analysis cannot solely focus on hardware investment. The pure vision solution requires extremely high computing power, necessitating powerful AI chips and complex neural network models. Tesla’s self-developed Dojo supercomputer and FSD chip also involve significant R&D investment. In contrast, while the multi-sensor fusion solution has higher hardware costs, it has relatively lower computing power requirements, allowing for faster commercialization.
Safety Performance: Redundancy Assurance vs. Intelligent Response
In terms of safety, the two camps hold opposing views. The lidar camp insists that lidar is the only perception device that can independently model without visual information, making it the most direct method to enhance safety redundancy. Jin Yuzhi, CEO of Huawei’s Intelligent Automotive Solutions BU, stated that the pure vision solution has inherent shortcomings and requires learning to perceive obstacles, while lidar can directly perceive obstacles without learning.
The pure vision camp emphasizes the advantages of “computing power + models.” XPeng Motors stated that the clarity of cameras has now surpassed that of the human eye, and as long as the computing power is sufficient, the response time can be several times or even ten times faster than that of fusion solutions. They believe that the significant improvement in AI model and chip performance is sufficient to support high frame rates and low latency real-time perception, which can react even faster than lidar in handling sudden situations.
Market Outlook: Differentiated Development Paths
From the current market landscape, it is likely that both technical routes will coexist for a long time, each occupying different niche markets. The pure vision solution, with its cost advantage, is more suitable for the L2-L3 level passenger car market; while the multi-sensor fusion solution is better suited for the L4-L5 level commercial autonomous driving fields, such as Robotaxi and logistics transportation.
It is worth noting that these two routes are not completely distinct but are learning from and integrating with each other. The pure vision solution is also incorporating more sensors; in multi-sensor fusion solutions, the role of visual algorithms is becoming increasingly important, serving as a key to understanding scene semantics. In the future, as technology advances and costs decrease, both solutions may find a new balance.
The ultimate winner of this technical route controversy is likely not a single solution, but rather companies that can flexibly choose the optimal solution based on different application scenarios. The future of autonomous driving may not lie in an “either-or” choice, but in the wisdom of “adapting to needs.”