How will intelligent driving continue to iterate?

The evolution of intelligent driving technology depends on three key elements: algorithms, computing power and data. There is gradually a consensus in the industry that intelligent driving applications have some characteristics:

● As enterprises continue to try, the algorithms of intelligent driving tend to be unified, led by BEV+Transformer. Especially after the large model has run through in all aspects, this belief has been further deepened.

● In terms of computing power, with the demand for E2E, car companies need stronger computing power chips to support algorithm iterations. Orin is no longer enough, and everyone is preparing to use Thor. (Previously, Huang Jiaozhu’s release did not feel special, but now start to make sense)

As for the data fed back by mass-produced cars, as Tesla claims that it can cover Corner Place with the generated speed, the weight of the data feedback may have been reduced! Elon Musk said: "Tesla has been able to generate real-world videos with accurate physics for about a year. It's not particularly interesting because all the training data comes from cars, so it looks just like Tesla videos, albeit a dynamically generated (rather than memorized) world.”

Here, the biggest change is the data element of intelligent driving. Before Openai released Sora, everyone believed that data was fundamental. In intelligent driving scenarios, the key factors affecting data generation and iteration mainly include the following:

● The number of mass-produced cars, that is, enough data sensors are needed to start working;

● The efficiency and quality of data sensors, standard sensor solutions have a stronger role in promoting technology iteration;

● Free-to-use smart driving products facilitate data feedback and technology iteration.

As the scale of lidar mass production increases, the cost decreases, which is conducive to large-scale application and promotes the iteration of more intelligent driving products to higher levels. But with the advancement of large model technology and the rapid development of visual technology, lidar is more like a redundancy.

More and more L4 players are launching smart driving products suitable for passenger cars. The application of fully autonomous driving technology can accelerate the launch of high-tech smart driving products. However, these players do not have a strong sense of presence and cannot implement cost-effective solutions.

Sensor solutions with the same specifications can improve the iteration efficiency of intelligent driving capabilities. The smaller the number of sensor solutions and the higher the level of achieved capabilities, the stronger the technical capabilities of intelligent driving.

Zhi Neng Comments: Does the picture below represent your opinion? You are welcome to leave your thoughts in the message area!

China's smart driving market shows an obvious echelon pattern. Manufacturers that can achieve NOA in urban areas are considered to be the leading generation, while manufacturers that can achieve high-speed NOA are regarded as the current generation, while manufacturers that lack NOA functions are directly regarded as the backward generation. . Various manufacturers are gradually forming unified standardized data containers, which is conducive to the collection and utilization of data and helps intelligent driving systems develop to higher-level iterations.