한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

as the leader of "end-to-end", mr. mao jiming pointed out that the cost structure of intelligent driving solutions will also change from the traditional modular architecture to the end-to-end model. a large number of rule writers will migrate to the data aspect, which is a good thing for oems with mass production capabilities. due to the low cost of acquiring data, the overall cost of intelligent driving solutions will actually drop significantly further. mr. lou tiancheng believes that in the short term, purchasing high-computing power chips will indeed increase current costs, but in the long run, once end-to-end technology is mature and applied, the initial investment costs will gradually be diluted.

however, this does not mean that technical challenges have completely disappeared. to effectively train an end-to-end model, a large amount of computing resources is required. mr. wang panqu said that the training cost is the biggest challenge. if it is just a simple end-to-end autonomous driving model training, hundreds of high-computing gpus are enough. but to invest in the long term and ensure end-to-end quality, the training computing power scale of each autonomous driving company will basically reach the kilocalorie level, and car companies will also invest more.

mao jiming gave a more specific end-to-end computing power requirement: the entire system requires at least two nvidia orins or a single nvidia thor. he said that the computing power requirement of a pure end-to-end system is less than the total computing power requirement of a modular architecture, but in addition to the main system, mass-produced end-to-end systems often have a bypass system, and its computing power requirement is generally equivalent to that of the previous modular architecture. however, wang panqu believes that with the increase in the capabilities of vehicle-side computing chips, computing power will not become an obstacle to future end-to-end vehicle deployment.

lou tiancheng holds the same view, saying that the total amount of code will be significantly reduced from the classic architecture to the end-to-end architecture, and the computing resource consumption brought by the end-to-end neural network will not necessarily be significantly improved compared to the bev model. "the desire for higher computing power comes more from the increase in the number of model parameters and model performance, rather than from the end-to-end transformation." he pointed out that from the perspective of end-to-end landing applications, relevant companies should think more about how to make full use of existing chip computing resources to improve utilization efficiency.

in short, the "end-to-end" revolution is changing the autonomous driving industry. it is not only a technological innovation, but also a new development direction and opportunity, and it also brings new challenges. in the future, with the continuous advancement of technology and the improvement of computing power, we will see more complete and more intelligent autonomous driving systems, bringing more convenience and safety to our travel.