Rebirth: Starting from Lighting Up the Tech Tree

Chapter 142 Intelligent Driving

Two weeks later, Shen Yiming printed out a thick stack of technical roadmap reports and placed them in front of Zuo Cheng.

Zuo Cheng opened to the first page, which read: "Ground Autonomous Driving Migration Scheme Based on UAV Multi-Sensor Fusion System." Below the cover was a line of smaller print, handwritten by Shen Yiming: "First draft, likely to be criticized."

Zuo Cheng didn't scold him. He flipped through the report from beginning to end, only folding a corner on the last page.

"Tell me about your chosen technology approach." Zuo Cheng closed the report and pushed it to the table.

Shen Yiming pulled out a chair, sat down, opened his laptop, and switched to a comparison image.

"Currently, there are three main paths in the industry. The first is pure vision, represented by Tesla, which relies on cameras and neural networks. It's low-cost, but accuracy drops significantly in strong light, low light, and rainy conditions. The second is the LiDAR route, which most domestic companies are pursuing. It offers high precision, but the unit price of LiDAR is 20,000 to 30,000 yuan, which is prohibitive given the overall cost of a vehicle. The third path is what we're taking."

"you say."

"The approach involves a fusion of three sensors: vision, millimeter-wave radar, and low-line-count laser radar. Vision processes texture and color, millimeter-wave radar provides rain and fog penetration and velocity measurements, and laser radar supplements depth information. This three-way data fusion is performed in real time. We've already implemented a similar architecture on drones; the core logic is the same."

Zuo Cheng nodded, signaling him to continue.

"We have a foundation in the perception layer; the problem lies in the decision-making layer. The decision-making environment for drones is relatively simple, mainly dealing with other aircraft and ground obstacles in three-dimensional space. The ground environment is far more complex, involving pedestrians, non-motorized vehicles, traffic lights, intersection logic, overtaking, and lane changing—all these scenarios require complete retraining."

Zuo Cheng didn't rush to reply, letting Shen Yiming finish speaking.

"But there's something we can use directly." Shen Yiming switched to a densely packed data distribution map, "The scene annotation system, annotation tools, and data quality review process we've accumulated on drones—this infrastructure can be directly reused. The most expensive part of intelligent driving isn't the algorithm, but high-quality data annotation, and we already have a mature pipeline for that."

"This is something I didn't expect." Zuo Cheng looked at it for two seconds and then wrote it down beside him.

Where does the data come from?

Shen Yiming switched to the next page, which displayed a timeline: "Two steps. First, we'll use simulators to run synthetic data, building the basic structure of the decision-making model. Second, we'll collect real-vehicle data from automakers and iterate as we run. We'll build the simulator ourselves, which will take about three to four weeks, keeping costs under control. Additionally, we have high-precision street-level map data of the urban areas the drones have flown over, which is very valuable for initial training."

Zuo Cheng looked at the page in Shen Yiming's report that had been folded over—the section on V2X communication. He picked up the report, flipped back to that point, and pointed to a parameter: "Here, you wrote 50 milliseconds for vehicle-to-vehicle communication latency. Why not use our Nebula protocol's 3-millisecond version?"

Shen Yiming paused for a moment, then realized, "I didn't think that way at the time. The Nebula Protocol was designed for the Sky Dome satellites; its compatibility with ground-based systems wasn't assessed."

"Let's assess this; this is our differentiator. If we incorporate 3 milliseconds of vehicle-to-vehicle communication, in an emergency stop scenario, the braking signal from the vehicle in front can reach the vehicle behind 50 milliseconds in advance, directly reducing the probability of an accident by an order of magnitude." Zuo Cheng put down the report. "If this can be implemented, the technological approach will have a real barrier to entry."

Shen Yiming quickly jotted down some notes on the scrap paper beside him, then looked up: "Has Fang Ze's timeline for automotive certification been released yet?"

"It's out." Zuo Cheng pushed another document over. "Functional safety ISO 26262, we're aiming for ASIL-B level, one level lower than the highest, but sufficient to cover the passenger car market. The timeline is at least 18 months. Fang Ze suggests proceeding simultaneously, developing and testing at the same time, overlapping the time windows."

"18 months, that's roughly the middle of next year," Shen Yiming calculated in his mind. "If we can have a technology demonstration version by the end of this year, and finalize cooperation with car companies in the first half of next year, we can just meet the certification deadline."

Zuo Cheng leaned back in his chair, his tone calm: "Right now, the most important thing isn't certification, but to quickly produce a demo system that car manufacturers can see. It doesn't need to be at mass production level, but it needs to be functional and demonstrate the depth of our technology."

"for how long?"

"Three months."

Shen Yiming paused for two seconds, mentally reviewing the number, then nodded: "I need to hire two more people from outside, in the field of autonomous driving, with prior experience in LiDAR perception or control algorithms."

"Chen Hao will give us the green light," Zuo Cheng said decisively. "But there's one condition: the core algorithm team must remain within 402; it cannot be outsourced."

"clear."

After the two finished discussing the technical approach, it was already getting dark outside the window. There were still people in the office corridor, the sound of keyboards clicking incessantly.

Han Lu knocked twice on the door and poked her head in: "President Zuo, the car company has replied. The CEO is willing to meet, and it's scheduled for next Wednesday."

Zuo Cheng said, "Okay."

Han Lu added, "That CEO is named Wei Dongsheng, in his fifties, and has been in the automotive industry for twenty years. His intelligent vehicle team started developing it independently three years ago, but hasn't succeeded yet, and is currently looking for external partners. I've asked people who have worked with him; he's straightforward, doesn't like beating around the bush, has some technical knowledge, and makes decisions quickly."

Zuo Cheng was satisfied with this description and said, "That makes things easier to discuss."

Han Lu left, and Shen Yiming packed up his laptop and prepared to leave. He stopped at the door, turned back, and said, "President Zuo, regarding that 3-millisecond latency V2X communication, if the Nebula protocol can be adapted to the ground end, our technology might really be a generation ahead of Tesla."

"It's not a possibility," Zuo Cheng said calmly, "it's certain."

Shen Yiming smiled, pushed open the door, and walked out.

Zuo Cheng sat back down at the table, unfolded a sheet of white paper, and wrote a milestone in the middle: Three months later, technical demonstration.

Then he wrote a name next to it: Wei Dongsheng.

He held the pen upright and tapped it lightly on the table a few times. In his previous life, CEFC had bet on the intelligent driving race, but Lin Jianhua was too conservative, clinging to a hardware-centric mindset. They burned through over a billion yuan in three years, ultimately veering off course and producing a hybrid product that became a laughingstock in the industry. In this life, Zuo Cheng didn't intend to repeat the same mistakes. He had already chosen his technological path and figured out the cooperation model. All that was missing was the official starting point.

Next Wednesday will be where this road truly begins.