At this moment, the technical lead spoke again.

"If this type of equipment is mass-produced and popularized on a large scale, the value of all existing air defense tactics and squad confrontation training will be greatly discounted."

"Because current mainstream countermeasures are all built on the premise that drones rely on communication links.

Therefore, electronic jamming and signal suppression can deal with remote-controlled models.

But they have little effect on fully autonomous aircraft like ours.

Even if brute-force electromagnetic pulse equipment can be used to damage the hardware,

such equipment is huge in size and expensive to manufacture, making it an easy priority target on the battlefield."

"And there is an even bigger problem.

That is detection and cost.

Drones like ours can maintain ultra-low-altitude flight of just a few meters.

With the help of buildings and trees to conceal their movements, it is difficult for radar to capture them in advance.

By the time they discover the target, our drones can basically be said to have already arrived right above their heads.

Moreover, the mass production cost of a single unit is relatively not too high.

Yet equipment capable of stably intercepting it easily costs tens of millions.

The asymmetric confrontation brought by such a cost gap can be said to be very difficult to resolve."

...

Afterwards, everyone naturally continued the discussion.

Even though small remote-controlled drones have already begun to be widely used in the market,

countries still do not have cost-effective countermeasures to this day.

Remote-controlled models are already difficult to guard against, and this kind of autonomous AI drone that requires no manual control is on a completely different level of threat.

Soon, an engineer proposed the subsequent iteration direction.

"Currently, our single-machine autonomous combat has been verified as a feasible solution.

Then I think we can promote two routes simultaneously in the future.

First is to develop a small fixed-wing platform adapted to this intelligent system to improve range and flight speed.

Second is to break through multi-machine networking technology and try to build a collaborative combat drone swarm."

Hearing this, Jiang Huaixing also nodded in recognition.

You must know that the other party's conception was precisely the direction of Jiang Huaixing's long-term planning for the future.

Moreover, not long after the baixing drone came out, relevant institutions took notice of it.

Later, they saw the development potential of White Star Technology's AI technology.

Therefore, they took the initiative to propose mounting the intelligent algorithm on an unmanned combat platform.

After multiple rounds of demonstration, this time the project for this test was implemented.

At this moment, Jiang Huaixing finally spoke.

"Alright, then our next stage of research and development focus can be divided into three parts.

First, continue to promote the self-research of dedicated airborne computing power chips.

Strip away redundant modules of civilian chips, deeply optimize for multi-sensor parallel computing, and solve the computing power bottleneck from the root.

Second, travel to various wild locations to collect scene data, continuously train the recognition model, and reduce the probability of misjudgment.

Third, start the development of cluster collaborative algorithms to achieve multi-machine situational sharing and autonomous division of labor and cooperation."

...

After giving the instructions, the people present started discussing around these three directions again.

Listening to everyone's discussion, Jiang Huaixing's thoughts also began to wander.

This autonomous drone system cannot be developed in isolation.

Because the consumer terminal chip R&D team of Xinghan Semiconductor is currently iterating a new generation of AI computing units.

Their R&D goal is low power consumption and miniaturization of mobile-end chips.

This is actually highly consistent with the demand for airborne chips.

They can even completely share the underlying IP core.

After removing redundant circuits,

two product lines can be completely derived, thereby sharing the high cost of tape-out and R&D investment,

and realizing two-way interoperability of technical achievements.

Moreover, current consumer phones need edge AI to support imaging and intelligent interaction.

Meanwhile, unmanned aerial vehicles in the sky and smart mobile platforms on the ground require highly reliable airborne computing units.

Then if the underlying technology comes from the same source, a unified technical base can be built completely.

"Chip projects cannot be carried out behind closed doors."

Jiang Huaixing spoke again.

"We can connect the semiconductor R&D line with the unmanned system laboratory and jointly establish a special task force.

We can completely reuse the mature AI core architecture of the mobile terminal and develop an airborne version with targeted optimization.

In the long run, local edge AI computing is an inevitable trend.

Whether it is battlefield equipment or consumer electronics, they will heavily rely on this technology in the future."

Hearing Jiang Huaixing's words, everyone instantly understood the value within.

You must know that if the R&D of a chip architecture can simultaneously empower consumer business and cutting-edge unmanned equipment,

the return on investment ratio will directly increase significantly.

Soon they finalized the general plan.

At this time, Jiang Huaixing also directly made the final decision.

"The priority of this project continues to be raised, and special funds, test sites, and data resources will all be supplied with priority.

There is no need to rush for short-term benefits; we must steadily iterate and optimize the technical route.

As long as we guard this track, we can take the initiative in combat under complex environments in the future first."

...

After the test at the test site ended, all test flight recordings, data logs, and defect reports were sorted and archived overnight.

Early the next morning, they were transmitted back to the headquarters in Shenzhen at the first instance.

That scene at the test site featuring millimeter-level hovering obstacle avoidance and fully autonomous return to base—

everyone who has seen it knows it in their hearts.

This set of AI unmanned systems has basically run through at the algorithm level, and what is lacking now is the hardware.

At 10:00 AM, White Star Technology's special meeting on drones started on time.

The large screen in the meeting room was divided into two parts, simultaneously broadcasting two test flight videos of the drone.

On the left was the first-generation machine test in early March.

It can be clearly seen that this drone's algorithm was immature, its path was rigid, and its search efficiency was also very low.

When encountering complex scenarios, it was very easy to overload the computing power, and then it would crash straight into the wall and shut down.

The whole process exposed its fatal shortcomings in processing peak data.

On the right was yesterday's latest test.

In roughly the same airspace and under similar obstacle scenarios,

the drone autonomously locked and stopped the moment it was about to hit the wall, and then the lidar instantly reconstructed the environment.

After that, it smoothly returned to the starting point.

After the video ended,

the technical lead of the drone project stood up and reported:

"After 8 months of iterative upgrades,

currently we have completed the closed-loop of the entire logic of perception fusion, spatial modeling, path planning, and autonomous return."

"It can be said that on software-level issues,

we have basically optimized everything that can be optimized.

The only bottleneck currently trapping the performance of the whole machine and restricting actual combat implementation is the existing vehicle-mounted computing power platform."

Having said that, he clicked open the parameter comparison table and continued:

"The BXW2 chip we currently carry is designed for civilian smart vehicles.

Its working scenarios are regular roads, stable lighting, and smooth data input.

Even if we removed some redundant designs,

but..."