Chapter 217: Launch Vehicle

One month later.

Dojin Soft Seochon office.

Inside the glass-enclosed control area, the control panel of a prototype vehicle connected to server racks was glowing.

Jo Seong-hwan, Park Cheol-jin, and I stood in front of the large display mounted on the wall, staring at the screen with tense eyes.

“Ready, sir.”

As Han Yu-ju reported, the text “Luna Beta Version – Kernel 2.0” blinked on the display.

Today was the day to install the Luna Beta version on one EV-Nova and conduct a simulation test for booting and cognition.

After taking a deep breath, I spoke in a low voice,

“Proceed.”

As Han Yu-ju pressed the touchpad,

An operation signal was sent from the server, and as the Black Robin chipset and Luna inside the vehicle connected, the vehicle’s HUD began to glow slowly.

[Initializing…….]

[Hardware Check: Black Robin AI SoC, Neural Engine: Active]

[Sensor Array: Camera(6) - Connected]

[Vehicle Interface Protocol: CANbus, Ethernet - Stable]

[System Status: Nominal]

After confirming the boot was complete, I slowly opened my mouth,

“Luna, how’s your status?”

[All sensors functioning normally, processing unit load rate 15%]

A calm voice came through the speaker.

While there were still some unnatural aspects, it wasn’t a bad response for a test operation level.

Han Yu-ju added an explanation,

“The Black Robin chipset currently in use provides about 500TOPS of computational power per second, but to perform perfect autonomous driving and stable cloud-based feedback learning, the next-generation chipset needs to provide at least 2,000TOPS or more of computational power.”

TOPS measures the number of addition, multiplication, and other operations performed per second in trillions, indicating AI processing capability.

While 500TOPS might not seem impressive,

The BR1 (Black Robin 1) chipset had at least five times better performance than the JW2 chipset used by Tesla,

And its performance was far superior to the 144TOPS level chipset of the JW3 scheduled for mass production next year.

Chilbong, who was developing the Black Robin 2 chipset, added,

“The BR2 currently under development is designed to achieve over 2,000 TOPS per second. Structurally, the key is a neural engine architecture with more than four times the parallel computation units compared to the existing one. To implement this, the hardware kernel of LUNA-EVX needs to be much more stable than it is now.”

Park Cheol-jin turned his head,

“Isn’t the LUNA-EVX equipment upgrade test already finished?”

“The first verification is done, but two more things are needed to apply it to the actual design.”

Chilbong held up two fingers,

“One is an L3 cache bus structure optimized for AI inference processing, and the other is a synapse matrix computation engine for physical AI.”

“Synapse matrix?”

Park Cheol-jin asked again,

“Existing neural computation is simply the repetition of vector and matrix multiplication, but to extend Luna’s emotional and contextual judgment functions, it requires a synaptic structure mimicking human neurons…….”

Chilbong, who was explaining, suddenly paused.

He realized that most of the people present couldn’t understand what he was saying, despite his efforts.

“Anyway…… It will take at least another year or two for BR2 to be developed.”

Since I already knew this, I didn’t talk much about BR2,

“For now, even with Black Robin 1, there’s no problem achieving Level 3 or higher autonomous driving according to SAE standards.”

“However, at the current level, only basic AI cognition and route judgment are possible. To achieve Level 3 or higher autonomous driving as you mentioned, a connection to a cloud system capable of handling large-scale data processing is necessary.”

The data generated by individual vehicles is vast and heavy, so to collect, process, and redistribute it in real-time, not only a powerful cloud system but also

A network system to connect it is needed.

I turned my head to look at Director Kim Se-bin,

“Can we hear a report on the progress of Luna Link today?”

“Yes, we’ll have it ready for you after the test.”

Nodding, I turned back to Han Yu-ju as the formal cognitive test procedure began.

Shortly after, the words “Cognitive Simulation Test – Scenario 1” appeared on the HUD screen.

“The first scenario is recognition of static objects, road boundaries, signs, and crosswalk pedestrians.”

As soon as Han Yu-ju finished speaking, the front display projected EV-Nova entering a virtual driving environment.

A simulation space replicating a downtown alley.

On both sides were illegally parked vehicles, and on the sidewalk were scooters, trash cans, and streetlights placed irregularly.

[Visual Processing Active]

[Object Tagging: 61 Targets]

[Priority Classification: Class A – Pedestrian, Class B – Obstacle, Class C – Reference Marker]

On the top of the display, Luna’s computation screen showed real-time object recognition tags being generated and classifying surrounding objects.

“Pedestrian detection rate 97.4%, object clustering accuracy 93.8%.”

EV-Nova smoothly decelerated in front of a human model,

And at the next traffic light, it stopped while considering the distance to the virtual vehicle behind it.

Those watching, including Seong-hwan, gasped in amazement,

“Wow, for a beta version, this is incredible.”

“If this works in real life, we could quickly catch up to leading companies like Tesla and Google Waymo.”

I waved my finger from side to side,

“It looks good because it’s a prepared scenario. To handle ‘exception situations’ on real roads, this isn’t enough.”

“To address that, the network issue with the cluster needs to be resolved.”

Han Yu-ju brought up the network issue again,

“Kim Se-bin will explain the progress on that part soon, so let’s finish the test first.”

“Yes, then let’s start the second scenario right away.”

As Han Yu-ju pressed the button, the simulation environment on the display changed.

This time, it was a scenario with complex variables: rainy night, road construction zone, and sudden object appearance.

[Environmental Condition: Rain / Low Visibility / Wet Road Surface]

[Simulated Variable Input: 3 Events – Construction Zone, Obstructed Signage, Sudden Pedestrian Entry]

Luna quickly scanned the driving environment and recognized the silhouette of a worker in a fluorescent vest on the rain-soaked asphalt.

Then, it simultaneously analyzed the cones in the construction zone and the road signs blurred by rainwater.

[Dynamic Route Adjustment Initiated]

[Signal Interpretation Confidence: 89.3%]

[Object Reclassification: Pedestrian, Debris, Warning Sign]

[Real-time Path Correction Applied]

“Alternative route secured, decelerating.”

As Luna’s voice came through the speaker, EV-Nova changed lanes on the wet road without slipping,

And returned via a detour, avoiding obstacles.

Shortly after, a child mannequin suddenly appeared on the sidewalk.

Luna detected it within 0.8 seconds and decelerated,

Emitted a warning sound through the speaker, and activated the hazard lights.

[Emergency Braking Activated]

[Obstacle Velocity: 0.3 m/s – Non-human Object]

[Risk Factor: Low – Proceeding After Clearance]

“This detection rate and reaction speed are quite sophisticated for a beta.”

In response to Seong-hwan’s words, Han Yu-ju replied immediately,

“But you must understand that what you just saw is only one out of tens of thousands of situations.”

I added,

“The real problem is that these exception situations aren’t experienced by a single vehicle alone. Hundreds of thousands of vehicles in the city experience them simultaneously, and they need to share and learn from each other for true autonomous driving.”

“That’s why Luna Link is essential for this project.”

At Director Kim Se-bin’s words, I nodded,

And immediately moved to continue the discussion with a report on Luna Link.

Before Professor Kim Se-bin began reporting on the development status of Luna Link,

I briefly explained what Luna Link was and why it was necessary.

[Luna Link]

I wrote the name in Korean on the board in the corner of the meeting room,

“In short, Luna Link is a global-scale data neural network.”

Park Cheol-jin showed curiosity at my explanation,

“Data neural network?”

“Just as neurons in the human brain exchange information through neural networks, it refers to a network structure where the AI of millions of vehicles operating worldwide constantly interacts and shares learning data.”

I drew a simplified Earth on the board,

Then added models of vehicles, data centers, and communication satellites orbiting in space.

“In the future, the data collected by Luna’s individual vehicles will be vast and require real-time processing. Therefore, a ground network alone won’t be enough for seamless data flow.”

“So, a satellite-based communication network like Luna Link is necessary?”

“Exactly.”

Director Kim Se-bin stepped forward in front of the board and added an explanation,

“As the CEO mentioned, we need to deploy a large number of ultra-small communication satellites in low Earth orbit, 500 to 600 kilometers above the ground. Only then can we achieve ultra-low latency communication of under 20ms globally.”

Seong-hwan, who had been listening quietly, expressed doubt,

“How many satellites would that require?”

“At least 1,000 are needed, and to achieve perfect global coverage, we’ll ultimately need to operate about 5,000 low Earth orbit satellites.”

At the staggering number of 5,000, the faces of those attending the meeting stiffened slightly.

“With that many, the launch costs alone would be astronomical.”

The production cost of one launch vehicle averages 150 million dollars,

Which is nearly 200 billion won in our currency.

To launch 5,000 satellites, 5,000 launch vehicles would be needed.

Even if multiple satellites are launched at once, nearly a thousand launches would be required.

Seong-hwan exclaimed in shock,

“Then the launch costs alone would exceed 200 trillion won?”

It wasn’t an arithmetically incorrect statement.

However, there was, of course, a way to drastically reduce costs.