Today, we try to pry open the "black box" of autonomous driving.

On March 13, 2004, the first DARPA Grand Challenge — the world's first autonomous vehicle race — kicked off in the Mojave Desert: 106 teams entered, 15 made it to the finals, and exactly zero finished. Who would have guessed that this seemingly failed competition would become an indispensable chapter in the history of autonomous driving?

Nineteen years later, autonomous driving is no longer the wild dream of scientists and engineers. Since the first half of this year, more and more automakers have announced their city-by-city rollout plans, and L2-level autonomous driving features have started trickling down to vehicles priced under 100,000 RMB. Self-driving technology is increasingly becoming part of everyday life.

For this episode, we're joined by two guests: Hong Zexin, partner at CHEK (an automotive intelligence platform) and a top contributor on Zhihu for automotive and autonomous driving topics; and Zhou Lin, who spent eight years in the autonomous driving industry — including stints at Hesai Technology and Momenta — and now works as an intelligent driving product marketing expert at an automaker. Together with Yu Haonan from ZhenFund's investment team, they'll discuss the democratization trend in autonomous driving and the lack of industry consensus on intelligence standards.

In this episode, you'll hear about: Is it better to pre-install expensive hardware hoping to unlock more features through iterative updates, or to use cheaper hardware to get lower-level autonomous driving into more hands first? How do automakers currently conduct evaluations? How much should consumers pay for L2+ advanced driver assistance? How to choose the right smart car without unified standards? What's the difference between human driving and intelligent driving? And how do we attribute accidents caused by assisted driving features?

Host

Yu Haonan, Investment Analyst at ZhenFund

Guests

Zhou Lin — 8 years of autonomous driving experience; formerly Marketing Director at Hesai Technology, Market Strategy at Momenta; currently Intelligent Driving Product Marketing Expert at an automaker

Hong Zexin — Partner at CHEK, Intelligent Driving Product Expert, Top Contributor on Zhihu for Automotive and Autonomous Driving Topics

Timeline

03:41 Factors driving the wave of autonomous driving democratization

10:28 How much do consumers need to pay for L2+ autonomous driving?

14:22 The hardware pre-installation vs. low-cost-first route debate

17:35 Automakers may not have figured out which route to take

21:05 Why automakers conduct evaluations

23:12 Smart cars are increasingly becoming black boxes

26:03 Road testing vs. simulation platforms: two approaches to automaker evaluation

30:40 Good features depend on the real-world constraints faced

34:50 Finding your own standard: the right car requires hands-on experience

40:44 Finding value by starting from your own needs

42:25 Highways, seemingly mature, remain a weak link for assisted driving

45:23 The difference between human driving and intelligent driving

47:50 How to gather evidence and attribute accidents caused by assisted driving features

52:22 The industry calls for standardized evidence-gathering rules

56:21 With a budget above 300,000 RMB, which models would you recommend?

59:56 With a budget under 200,000 RMB, which models would you recommend?

Reference Materials

L2: Level 2 autonomous driving is a classification standard where the vehicle handles basic operations while the driver monitors surroundings and remains ready to take over at any time. Main functions include ACC (Adaptive Cruise Control), automatic following, and automatic parking. Per SAE (Society of Automotive Engineers) standards, autonomous driving spans six levels: L0 (no automation), L1 (driver assistance), L2 (partial automation), L3 (conditional automation), L4 (high automation), and L5 (full automation).

ACC (Adaptive Cruise Control): Building on traditional cruise control, ACC uses radar to detect relative distance and speed to vehicles ahead, actively controlling speed to maintain automatic following. The system can automatically switch between cruise control and following modes based on traffic ahead.

LCC (Lane Centering Control): A comfort-oriented assisted driving feature comprising Traffic Jam Assist (below 60 km/h) and Intelligent Cruise Assist (above 60 km/h). It monitors the vehicle's position relative to lane center and actively assists the driver in maintaining center-lane position, reducing steering burden.

Occupancy Network: An algorithm Tesla introduced in 2022 that divides the 3D world into grid cells, defining occupied vs. free cells. The concept uses "occupancy" rather than detection to display road information in real time, obtaining volumetric occupancy rates.

Orin: An autonomous driving chip from NVIDIA built on 7nm process technology, packing 17 billion transistors. The SoC integrates NVIDIA's next-generation GPU architecture, Arm Hercules CPU cores, and new deep learning and computer vision accelerators, delivering 7x performance improvement over the previous-generation Xavier.

AEB (Autonomous Emergency Braking): AEB systems use radar and cameras to measure distance to vehicles ahead or other obstacles. The control module compares measured distance against warning and safety thresholds, triggering alerts when below warning distance and initiating braking to stop the vehicle when below safety distance.

OTA (Over-The-Air technology): Technology for remotely managing firmware, data, and applications on vehicle component terminals via mobile networks (2G/3G/4G or WiFi).

Domain Controller: An intelligent computing platform for L3-L5 autonomous driving applications that integrates compute-intensive sensor data processing and sensor fusion with control strategy development into a single control unit, helping establish a structured and organized vehicle controller network.

End-to-End: Originally a deep learning concept; in autonomous driving, it refers to models that directly output driving behavior commands (steering angle, throttle, braking) from input images or video.

Production

Post-production: Keyone Studio

Contact Us

WeChat Official Account: ZhenFund (ID: zhenfund)

Listening platforms: Xiaoyuzhou | Apple Podcast | Ximalaya

Email: media@zhenfund.com

We welcome your feedback and suggestions in the comments!