Reborn in 2010, the AI ​​in my mind is too powerful.

When Aurora Group was founded, it was already a behemoth. At the end of last year, Chen Yiran made a major acquisition of several companies to form the foundation of Aurora Group.

After its establishment, it made numerous acquisitions in just over two months, and thanks to its ample cash flow, it expanded almost at a flat pace.

The group now has more than a dozen subsidiaries, spanning the United States, Japan, Germany, the United Kingdom, Canada, Denmark, France, and China, with dozens of work locations.

The number of employees has reached more than 20,000.

Of course, the newly acquired manufacturing companies make up the majority. Yaskawa alone has more than 10,000 employees distributed across more than a dozen factories worldwide. KUKA has more than 3,000 employees in Augsburg, iRobot has more than 500 employees in Boston, and Linac has more than 2,000 employees in Denmark.

These companies possess invaluable data and experience, but they are all isolated entities.

Yaskawa uses JIS standards for motor test data, KUKA uses DIN standards for simulation data, and DJI uses a custom format. Each company has a different R&D process. When only two or three companies collaborate, it's not a big deal, but now that a dozen or so companies need to confirm the design, it becomes a major problem.

Therefore, a unified R&D platform is essential.

Chen Yiran named the project the SEELE R&D platform, which is the German word for "soul".

This is considered the most important task for Aurora Group in the near future, and Chen Yiran directly assigned it to Gao Dewei to oversee the overall situation.

Of course, the most important part, including the basic structure, is almost complete; it just needs people to fill it in.

Core personnel need to be interdisciplinary talents; they need to understand deep learning and also have some knowledge of industrial fields.

Such people were extremely rare in 2012. Those in academia who were doing deep learning could produce a lot of papers, but they were completely out of their depth when it came to optimizing motor designs. Conversely, engineers from Yaskawa and KUKA, who had worked in their respective fields for over a decade or two, would have to start from scratch to understand concepts like gradient descent and backpropagation.

Fortunately, the company is big enough now, so we can always find talent. There are quite a few engineers who have studied deep learning in their spare time. In addition, FlowEDA has two long-term employees with undergraduate degrees in engineering.

We've finally gathered all the people in this field.

Then each branch office also needs to send people to cooperate. These people do not need to understand Deepflow, but they need to know their specific R&D needs.

A week later, the elites transferred from various subsidiaries arrived in Silicon Valley one after another.

The project team consisted of over twenty engineers of various types, plus six underlying architecture engineers from FlowEDA.

Chen Yiran returned to the company's conference room after a long absence and began setting goals for the project.

"I know that you are all elites in your respective fields. This project requires your close cooperation. The SEELE R&D platform will be the most important internal R&D tool for Aurora Group in the future."

Simply put, our main goal is to achieve three objectives.

The first is data integration. Each company uses a different data format, so the platform needs to create a data adaptation layer to convert the data from each company into a unified format and store it in the same data lake. In the future, when Yaskawa's motor test data is uploaded, KUKA engineers can directly see the corresponding simulation results without having to send and receive Excel files.

The second feature is model sharing. The platform provides a model repository where companies upload their trained models, along with complete parameter descriptions and test results. Other companies can then download, fine-tune, and deploy these models. For example, DJI can use Longyan's visual model, and Universal Robots can use Yaskawa's motor control algorithm, eliminating the need to reinvent the wheel.

The third is computing power scheduling. Deep learning training requires a large amount of GPU computing power, and it's impossible for every company to build its own cluster. The platform needs a unified computing power scheduling system. Whoever needs training applies for resources, the system automatically allocates GPU nodes, and automatically reclaims them after training is complete.

The reactions of the people below varied. Companies like FlowEDA, Wanguoyuan, and Uniview Technologies were all very familiar with Deepflow and its related operations.

The representatives of the newly acquired companies mostly had a clear look of stupidity in their eyes.

Fortunately, they don't need to do the specific things; they just need to make requests.

After assigning tasks, Chen Yiran returned to his hands-off manager mode, only occasionally appearing on the message board to resolve some technical issues they couldn't handle.

As May drew to a close, Chen Yiran visited SpaceX once again, this time not as a visitor, but as a shareholder to observe the Dragon spacecraft mission.

In the early hours of May 22, Falcon 9 took off from Cape Canaveral.

This is SpaceX's third launch of the Falcon 9 rocket and the second orbital flight of the Dragon spacecraft.

However, the first launch attempt five days ago was automatically aborted 0.5 seconds before ignition due to excessive pressure in the combustion chamber of one of the Merlin engines, prompting the computer to shut down.

Chen Yiran was watching the live broadcast in Silicon Valley. When the launch abort notification popped up on the screen, his first thought wasn't worry. He knew that SpaceX had succeeded in the end, so this should be just a minor setback.

Sure enough, after rearranging the equipment, the second ignition went smoothly.

The real big day is three days from now, on May 25th, when the Dragon spacecraft will be captured by the Canada 2 robotic arm of the International Space Station and docked in the Harmony node module.

This is the first privately owned spacecraft in human history to reach the International Space Station. Since the retirement of the Space Shuttle last year, the United States has lost its independent manned spaceflight capability and even its cargo transport has been disrupted.

If the Dragon spacecraft succeeds today, SpaceX will automatically receive a contract from NASA for the next twelve cargo missions.

On the same day, Chen Yiran arrived at SpaceX's current control center.

In the open-plan facility at SpaceX headquarters in Hawthorne, a huge projection wall is covered with real-time telemetry data, showing the Dragon spacecraft's position, speed, and distance from the space station.

The blue light from dozens of monitors illuminated the faces of the engineers in the rows below, and video signals from NASA were also being received. The mission control center in Houston was simultaneously monitoring the same set of data streams.

Chen Yiran secretly glanced at Musk beside her and sensed that he was actually quite worried.

Since smoking is not allowed in the control center, Musk had to keep a cigarette in his mouth the whole time, swaying left and right as if he had ADHD.

The docking process is very complicated. The Dragon spacecraft first automatically flew to a checkpoint 10 kilometers below the space station to confirm that the navigation system was working properly, and then gradually shortened the distance.

Once within two kilometers, the speed slows down to approach, and each step requires confirmation from both NASA and SpaceX before continuing.

Instead of heading towards the berth on its own, the Dragon spacecraft hovers ten meters away, its robotic arm extends to grab it, and then tows the spacecraft to the node module's berth. Safety redundancy is maximized; it's better to have the robotic arm do more work than to allow the spacecraft to autonomously approach the space station.

The operator of Canada 2 robotic arm controlled the arm remotely from inside the space station's dome module. The arm extended seventeen meters, and the lasso catcher at its end was aimed at the grabbing component on the belly of the Dragon spacecraft.

About three or four parameters on the telemetry data panel kept changing, the distance jumping from ten meters to five meters, 3.2, 2.1, 1.8. Then it stopped. No one spoke in the entire control center, only the sound of data refreshes and occasional short intercom messages.

Finally, the end effector of the robotic arm moved slightly, and the lock indicator light on the monitor turned green.

The control center erupted instantly. Some people stood up and slammed their fists on the table, some took off their headphones and leaned back, and many more threw their documents into the air—it was a scene reminiscent of the end of the college entrance exams. Musk, meanwhile, took the cigarette he'd been holding for almost forty minutes out of his mouth and let out a long sigh.

Next came a NASA walkie-talkie voice message: the Harmony node berthing hatch has been successfully unlocked, and the second row of locking bolts is in place. The signal stream showed that the Dragon spacecraft and the space station have completed their physical connection, and the solar panels have begun to rotate and adjust their attitude.

Musk and Chen Yiran gestured to each other, then Musk went to the front and started to motivate the employees.

After all, this success is extremely important, as it proves that SpaceX is no longer a company that needs continuous financial support.

Chen Yiran did not stay at SpaceX for long; he took his team directly to Norway.