Recently, autonomous driving has been catalyzed by intensive policies. According to Securities Daily, the pilot project of &#34;driverless vehicles&#34; for autonomous driving has been officially launched in Beijing. Pengpai News reported that Baidu Apollo and two other companies have obtained the first batch of test licenses for autonomous vehicles in Pudong New Area, marking the beginning of a new journey for driverless vehicles.
In the broader passenger car sector, domestic automakers have successively initiated the internal testing of &#34;urban NOA&#34; (Navigation on Autopilot) and are expected to develop a nationwide &#34;commuting NOA&#34; in the second half of the year.
In response to this, Westone Electronics stated in a report last week that the advent of autonomous driving is a turning point, and 2023 will be the breakthrough year for advanced driver assistance systems, which will accelerate their penetration.
Among them, computing power is the core for automakers to develop autonomous driving capabilities, and the emergence of large-scale AI algorithms such as Transformers is driving the transformation of autonomous driving algorithms. Open Securities stated in a report on July 9th that companies that can adapt Transformer algorithms well and help them land on in-vehicle platforms are expected to gain an advantage.
<h2>The Advent of Autonomous Driving Turning Point</h2>
Westone Electronics pointed out that the advent of autonomous driving is a turning point, and 2023 will be the breakthrough year for advanced driver assistance systems.
<blockquote>
NOA (Navigation on Autopilot) can automatically enter or exit highways during navigation. 2023 will be the breakthrough year for urban NOA 0-1, and its penetration will accelerate in the future.
</blockquote>
In addition, in the second half of 2023, major cities in China are expected to be the first to introduce Level 3 regulations, driving the rapid development of advanced autonomous driving (Level 3 and above). Safety is the core of advanced driver assistance systems (Level 3 and above), and LiDAR remains essential.
<h2>Automakers Compete in Computing Power</h2>
Computing power is the core for developing autonomous driving capabilities, and automakers have initiated a &#34;competition&#34; in computing power.
According to Open Securities, autonomous driving algorithms usually include perception, prediction, and planning modules, as well as auxiliary modules such as maps and positioning, to achieve the implementation of autonomous driving functions. The emergence of large-scale AI algorithms is driving the transformation of autonomous driving algorithms.
<blockquote>
In the perception aspect, Tesla, as well as domestic companies such as Xiaopeng and Li Auto, have already used Transformer+BEV and Occupancy Network to achieve real-time obstacle recognition and motion target detection, gradually breaking free from the limitations of high-definition maps, enabling urban NOA to quickly expand in cities without high-definition maps;
In the decision-making aspect, rule-based algorithms are still dominant, but by introducing reinforcement learning with scoring mechanisms, the decision-making module has certain learning capabilities and can learn from the experience of human drivers as much as possible;
In the control aspect, data-driven AI-based control algorithms are increasingly coming to the forefront. AI models can process driving behaviors in a smoother and more &#34;human-like&#34; manner, with strong generalization ability and improved adaptability to complex scenarios.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/df6e0e2b-071b-40a3-b19f-77b602d6941f.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/df6e0e2b-071b-40a3-b19f-77b602d6941f.png"/>
Furthermore, the revolution of autonomous driving algorithms is driving a competition in cloud computing power, while the landscape of edge chips is expected to be reshaped.
</blockquote>
1. The revolution of autonomous driving algorithms drives a competition in cloud computing power
Open Securities pointed out:
<blockquote>
A large amount of data processing, training, automatic annotation, simulation, and other work needs to be completed. Computing power is the core of automakers' autonomous driving capabilities, determining the efficiency and upper limit of algorithm iteration.
</blockquote>
Tesla stated that its total computing power will reach 100EFlops by 2024. Leading players in China are also not far behind, actively building their own data centers to pave the way for algorithm and data iteration and accumulation.
<blockquote>
In August 2022, XPeng Motors and Alibaba Cloud jointly announced the establishment of the &#34;Fuyao&#34; Autonomous Driving Intelligent Computing Center in Ulanqab, Inner Mongolia. In January 2023, Geely Auto also partnered with Alibaba Cloud to launch the Geely Xingrui Intelligent Computing Center. In 2023, Li Auto, in collaboration with Volcano Engine, also established an intelligent computing center in Shanxi. NIO and many other automakers are also actively promoting the construction of their own or cloud computing power.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d1399961-58ab-4629-b83c-528ebc9b3322.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d1399961-58ab-4629-b83c-528ebc9b3322.png"/>
</blockquote>
2. Chip adaptability to Transformer will affect future competition landscape
Open Securities pointed out that large-scale Transformers pose new requirements for chip architecture, and chip adaptability to Transformers will affect the competition landscape of future chips.
<blockquote>
Chip architecture determines the efficiency of algorithm execution. In recent years, with the popularity of Transformer algorithms in the AI field, led by Tesla, Transformers have been widely used in the autonomous driving industry. Compared with traditional Convolutional Neural Network (CNN) architecture, Transformer algorithms have higher computational requirements, higher precision, and higher operator complexity.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/9bbeb586-1b23-4329-af67-de7b3954f2d8.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/9bbeb586-1b23-4329-af67-de7b3954f2d8.png"/>
</blockquote>
Given the above characteristics, many chip manufacturers have launched chip products that can accelerate Transformers. Open Securities believes that:
<blockquote>
In the future, companies that can adapt well to Transformer algorithms and help them land on in-vehicle platforms are expected to gain an advantage. As the autonomous driving industry accelerates its maturity, companies in the supporting ecosystem will benefit greatly from the widespread acceptance of BEV+Transformer large-scale model autonomous driving algorithm construction.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4c984dfa-7183-4cb1-8915-67b22fab8a2e.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4c984dfa-7183-4cb1-8915-67b22fab8a2e.png"/>
Overall, Open Securities believes that in 2023, we will gradually see major car companies launching their own city-assisted driving models. The industry will present a diverse and flourishing state, undoubtedly contributing to the prosperity of the entire autonomous driving industry chain.
</blockquote>

Open Source Securities stated that the transformation of autonomous driving algorithms has driven a competition in cloud computing power. Major automakers are actively building their own data centers. Companies that can effectively adapt the Transformer algorithm and facilitate its implementation on in-vehicle platforms are expected to gain a competitive advantage.

Has the era of autonomous driving arrived? Breaking free from high-precision maps, car companies are also competing for "computing power"!