Rebirth: Starting from Lighting Up the Tech Tree

Chapter 143 Contact with Automakers

Wei Dongsheng was more direct than Zuo Cheng had imagined.

The meeting took place in the conference room of a domestic car company's headquarters. It wasn't a formal setting with carpets, but rather an open-plan studio. Several scale models of cars sat on the table, and the walls were covered with technical roadmaps. Wei Dongsheng, in his fifties, had a receding hairline, wore an unbuttoned suit, and gripped the handshake firmly.

He got straight to the point: "Mr. Zuo, I've reviewed your materials. To be honest, last month I met with three intelligent driving companies. Some had secured several rounds of funding, and some claimed to have already reached Level 4, but none of them made me feel they were worth discussing in depth. Your materials are the first one that makes me think the person who wrote the proposal understands cars."

Zuo Cheng didn't rush to answer, but first asked, "What do you think you understand?"

Wei Dongsheng raised a finger: "In your sensor fusion solution, the range parameters for the millimeter-wave radar are based on the 77GHz band, covering over 150 meters. Some solutions in the industry use 24GHz, which is cheaper, but the range is only 50 meters, making it impossible to brake in case of emergencies on the highway. You didn't mention this detail, but the data you selected is correct."

Zuo Cheng said, "President Wei is very observant."

"Having worked in the automotive industry for twenty years, it's no wonder I'm so meticulous." Wei Dongsheng pushed aside his water glass and cut straight to the heart of the matter, "Our own intelligent vehicle team has been working on it for three years. We've developed L2-level technology, but L2+ is stuck at the perception fusion layer. The problem lies in rainy and nighttime scenarios, where the accuracy drops below 75%. Do you have a solution?"

Zuo Cheng opened his tablet and pulled up a chart: "This is our drone's perception data during nighttime flights, in scenes with light levels below 30 lux. You can see the detection rate; it's consistently above 93%. The light distribution on the ground differs from that in the air, but our fusion algorithm's logic is universal for handling low signal-to-noise ratio inputs. It can be adapted."

Wei Dongsheng stared at the image for over ten seconds without saying a word. The vice president of technology next to him whispered something in his ear, and Wei Dongsheng waved his hand, signaling him not to interrupt and to continue looking at the image.

Zuo Cheng didn't urge him, nor did he try to fill the silence with more words. The data was there for them to see for themselves; that was more convincing than any explanation.

"How do you classify targets in low light?" Wei Dongsheng looked up and asked directly about the technical details. "The camera is basically useless at this point, how do you maintain this accuracy?"

Shen Yiming continued, "It's a weighted fusion of multiple sensors. In low light, the weight of the camera is reduced, while the weights of the millimeter-wave radar and laser radar are automatically increased. The current lighting environment is determined by an adaptive weighting network that outputs the confidence coefficients of each sensor in real time, and then dynamically allocates the fusion weights based on these coefficients. This network was trained in a drone nighttime inspection scenario."

Wei Dongsheng rested his elbows on the table, tilted his chin slightly forward, and after listening, said, "Adaptive weights, we've thought about that too, but we haven't been able to implement it successfully in practice. There's latency and jitter at transition boundaries, causing unstable perception results. Have you solved that?"

"It uses smooth interpolation with a hysteresis window." Shen Yiming took the tablet and pulled up another waveform graph. "Look at this curve. The weight switching has no abrupt changes; it's a continuous, gradual process. The hysteresis window is 150 milliseconds, which is exactly an order of magnitude larger than the sensor's frame interval, thus eliminating the jitter problem."

Wei Dongsheng and Vice President Bao exchanged a glance, and something changed in their eyes.

"You guys who make drones really understand sensor fusion," Wei Dongsheng said, his tone relaxing for the first time. "I've met several companies that make autonomous driving systems, and most of them only understand algorithms, not hardware. Hardware and software are separate. Your solution, however, integrates the algorithm and hardware features in a unified design, which is quite remarkable."

The CTO standing next to him, surnamed Bao, had been silent until now, but finally spoke up: "With what hardware configuration was this accuracy achieved?"

Shen Yiming continued, "Our self-developed perception fusion chip, combined with Star Technology's neuromorphic processor, keeps power consumption below 8 watts."

Vice President Bao raised an eyebrow: "8 watts? Our current solution is 25 watts, and the entire power supply design is based on that. How much can we reduce the hardware cost with 8 watts?"

"Optimizations on the algorithm side can reduce power consumption to this level, but achieving automotive-grade compatibility requires going through the certification process again." Shen Yiming didn't exaggerate, presenting the actual situation: "We've assessed that a complete ASIL-B certification takes approximately 18 months."

Wei Dongsheng spoke again: "18 months, that's the end of next year."

"This is the fastest path."

Wei Dongsheng leaned back in his chair, looked at Zuo Cheng, and changed his perspective: "If we provide the testing grounds and a test fleet, how long can you produce a road-ready demonstration version?"

Zuo Cheng calmly replied, "Three months."

Wei Dongsheng looked at him for two seconds. He didn't question the number; instead, he nodded, as if confirming some judgment.

"Okay, I'll give you three months," he said. "After three months, if you can produce even a prototype-level demo that can be looped in our testbed, then we can discuss the terms of cooperation."

The meeting ended at 3 p.m. Han Lu was waiting at the door. When she saw Zuo Cheng come out, she raised her eyebrows and asked with her eyes.

Zuo Cheng lowered his voice and said a single word: "Okay."

Han Lu breathed a sigh of relief and followed Zuo Cheng towards the parking lot. In the corridor, Vice President Bao caught up and called out to Shen Yiming. The two stopped and exchanged a few words in hushed tones. When Shen Yiming caught up, a hint of suppressed excitement lingered on his face.

"General Manager Bao asked me if we could show them the internal technical documentation on V2X first. I said I needed to consult General Manager Zuo."

"Send it to him," Zuo Cheng said directly. "Compile a list of the work we did on the Nebula Protocol ground-side adaptation. Don't hide the technical details; let them see our depth of expertise."

Shen Yiming was a little surprised: "Aren't you afraid of being plagiarized?"

Zuo Cheng pushed open the parking lot door, letting in the early spring breeze of Hangzhou: "Only what can be copied can be copied. With technology, those who know it can tell at a glance what level someone is at, but for those who don't, even giving them the complete source code is useless."

Shen Yiming thought for a moment and then remained silent.

On the way back, Zuo Cheng leaned against the window, looking at the city outside, while Han Lu was organizing meeting minutes in the passenger seat.

Three months. This was his promise, and also the team's bottom line.

He opened his phone and sent a message to Shen Yiming: "Finalize the recruitment requirements tonight and send them out tomorrow. Three months won't wait."