DJI claims urban smart driving breakthrough in China

On January 31, Zhuoyu, which was previously known as DJI Automotive, claimed a breakthrough with computing power for urban smart driving. The company claims that it can achieve an urban smart driving capability using computing power of just 32 TOPS on a 7V medium computing power platform.

Previously, most mainstream Chinese car companies that could achieve an urban smart driving capability used the Nvidia Orin X chip. Most, such as Xpeng and Li Auto, use two of the chips, whereas in the case of Nio, it uses 4. Each Orin X chip has 254 TOPS of computing power.

Zhuoyu has achieved this feat using its Chengxing Platform. The company uses the same end-to-end algorithm architecture on its 7V + 32 TOPS system as the model architecture used on the higher computing power 7V/10V + 100 TOPS system.

Zhuoyu claims to be the first smart driving supplier to realize end-to-end urban pilot driving assistance on a medium computing power platform. Thanks to already offering such a system for full power systems, Zhuoyu can realize crosscomputing power and cross-configuration reuse to a large extent.

The company claims that the improvement in experience brought about by the optimization of software and algorithms is more obvious than that of simply increasing the number of sensors. Using device-side computing power optimization technology, the model is jointly trained and optimized according to the characteristics of the chip. This enables end-to-end model deployment even when there is limited computing power.

- Advertisement -advanced

Zhuoyu’s Chengxing Platform is compatible with various chip platforms and has been used by several Chinese car producers. The company says it will provide a free upgrade service through software OTA to cars equipped with the basic version of the Chengxing platform, i.e., 7V + 32 TOPS, once the software reaches maturity and mass production. Presumably that will be achieved at some point during 2025.

Source: Fast Technology

- Advertisement -spot_img

Read more

Recommend News