Dongwu Securities: Comparative analysis of autonomous driving road tests for automotive companies in Q2 2024, Tesla's overall performance is outstanding

According to the information from Zhitong Finance APP, Dongwu Securities released a research report stating that under its attempt to build a smart driving road real test evaluation framework, Tesla (TSLA.US) performed excellently overall and achieved full-scene coverage. The trend of automotive AI intelligence transformation is clear, with algorithms as the backbone and industry trends well-defined. Downstream OEM players and midstream Tier suppliers are increasing their investment in automotive intelligence; the core link of smart driving, namely software + hardware + data, revolves around downstream OEMs, with data catalyzing algorithm efficiency and driving hardware iteration. Represented by Tesla, the application of algorithms towards full-stack end-to-end - world model continuous iteration, with functions being implemented. The team is optimistic about the leading smart driving companies and intelligent incremental components.

Algorithm/Function Theory Level:

Tesla's FSDv12 version continues to iterate, with perception-control full-stack end-to-end implementation supporting fully point-to-point navigation driving in North America. Domestic car companies use BEV+Transformer to achieve precise perception and gradually increase the proportion of learning-based control algorithms, with the overall algorithm framework trending towards comprehensive end-to-end; Huawei/Xpeng is expected to implement end-to-end smart driving algorithms in 24Q3, with a relatively advanced pace.

Actual Road Test Experience Level:

To make a real judgment on the strength and weakness of smart driving algorithms of different car companies, and to visually reflect the vertical iteration speed difference of car companies' smart driving capabilities, Dongwu Securities attempted to build a smart driving road real test evaluation framework: distinguishing between "start-stop + driving" dual links, with the former covering permission restrictions + start + takeover points + end in four parts, and the latter including city coverage (city opening situation) + road coverage (first type of takeover point) + temporary handling (second type of takeover point) in three parts, gradually approaching the boundaries of smart driving capabilities of different players. (PS: Smart driving algorithm capabilities and data accumulation form a positive cycle, with leading players iterating quickly, so the comparison of capabilities needs to be closely tracked at high frequency. This article only represents the comparison of capabilities at a certain point in time)

Car Company Dimension: Tesla performed excellently overall, able to achieve features such as FSD starting in place, unrestricted driving on all road sections, smooth handling in multiple scenarios, no forced handover of steering wheel, and automatic parking, coming close to human driving (slightly slower reaction speed). Leading domestic companies such as Huawei Aito/Xpeng have gaps in features like starting in place, road coverage, and scenario handling smoothness; other companies like IM Motors/Aita perform weaker overall in features like start-stop, road coverage, and scenario handling capabilities.

Scenario Dimension: Tesla has full-scene coverage, while leading domestic OEMs like Huawei/Xpeng have more takeover points. Apart from the prior issue of road coverage, Dongwu Securities distinguishes the proportion of excellent performance of different companies in scenes of varying difficulty: in difficult scenes, the rates of excellence for Tesla/Huawei Aito/Xpeng are over 90%/close to 40%/close to 40%; in simple scenes, the rates of excellence for Tesla/Huawei Aito/Xpeng are close to 90%/close to 50%/about 60%. (Huawei's weaker performance compared to Xpeng is mainly due to Xpeng's overall driving algorithm strategy being conservative, lacking efficient behaviors like active lane changing, while Huawei is more aggressive) Investment Advice: The trend of automotive AI intelligence transformation is inevitable, with algorithms as the backbone. We are optimistic about leading algorithm players maintaining a competitive edge.

The entire industry is accelerating its transformation towards intelligence, with clear industry trends. Downstream OEM players and midstream Tier suppliers are increasing their investments in automotive intelligence, indicating a clear trend. The core elements of intelligent driving [software + hardware + data] are all centered around downstream OEMs, with data catalyzing algorithm efficiency and driving hardware iteration. Represented by Tesla, the application of algorithms towards full-stack end-to-end - world model continuous iteration, with functions being implemented and realized.

OEM vehicle manufacturers & core chip hardware manufacturers & intelligent driving sensor manufacturers & independent algorithm companies are accelerating the layout of end-to-end algorithm development, driven by scenarios - data - cognition, continuously evolving. The development of intelligent driving algorithms has entered a challenging phase, requiring high investment to empower [large computing power + big data], in order to achieve the transition from L3 conditional automated driving to L4 fully automated driving.

Optimistic about leading intelligent driving companies and incremental intelligent components: 1) Huawei-related players [Changan Automobile + BAIC Blue Valley + Saic-Iss + JAC Motors]; 2) Leading new forces [XPeng Motors + Li Auto]; 3) Accelerating transformation [BYD + Geely Auto + SAIC Motor + Great Wall Motors + GAC Group]; 4) Core intelligent incremental components: domain controllers (Desay SV + Joyson Safety Systems + Huayang Group + Junsheng Electronics, etc.) + wire-controlled chassis (Benteler + Nidec + Top Group, etc.).

Risk Warning:

The iteration of intelligent driving-related technologies/policies may be lower than expected. If the pace of iteration of intelligent driving-related technologies is slower than expected, it may affect consumers' awareness and acceptance of intelligent driving, while a slower-than-expected pace of policy implementation may also impact the progress.

Leading car companies such as Huawei/XPeng may have lower-than-expected new car sales. If the sales performance of leading car companies in intelligent driving falls below expectations, it may hinder the increase in intelligent driving penetration, negatively affecting the sector