Das ist schrecklich! Es kann noch kriechen, auch wenn seine vier Beine abgesägt wurden. Der universelle Roboter-Hirn löst die "ununterbrochene" Ära der Roboter aus.

[New Introduction] The era of robots "breaking down" may finally come to an end! The general "brain" unveiled by Skild AI enables robots to keep moving forward even after suffering limb loss, damage, or even a complete body replacement. It seems to possess an indomitable survival instinct, with each injury turning into a new step forward. The flashing "Adapting…" on the screen seems to proclaim to the world that the truly unstoppable era of robots has arrived.

In science fiction, robots are omnipotent and can keep going even when damaged.

However, in reality, they are extremely fragile: a loose screw in a robotic arm can bring it to a halt, and a malfunctioning motor in a quadruped robot can make it "collapse."

A short video posted by Skild AI on X offers a new solution: the robot keeps moving forward despite the damage, with "Adapting…" constantly flashing on the screen.

The secret behind this is their brand - new concept: Skild Brain, an adaptable robotic "brain" independent of the body.

From Malfunction to Self - Adaptation: The Intelligent Aesthetics of Skild Brain

In our past imagination, the "brain" of a robot had to be closely attached to its body: every joint damage or servo failure could mean a complete system shutdown.

However, Skild AI has brought about a fundamental revolution - they proposed the concept of a "robot brain decoupled from the body," which is the core of Skild Brain.

Their vision is:

Omnipotent Form Intelligence: One brain for any task and any robot.

In this framework, Skild Brain is designed as a general - purpose intelligence capable of operating across different forms and tasks, and has been tested on various platforms such as quadruped, humanoid, and mobile robotic arms.

Details from their blog reveal that the control strategy is divided into two levels - high - frequency and low - frequency. The high - level is responsible for planning navigation actions, while the low - level converts the strategy into joint torques and angles, and can be migrated between different robots.

Even more astonishingly, the Skild team stated straightforwardly on X:

A severed limb? A jammed motor? As long as the robot can still move, this brain can make it keep going - even if it gets a brand - new body.

This statement is the manifesto of their system design: as long as there are available motion channels, the brain can find a way to act.

In the eyes of the outside world, this system is not just a technological showcase, but may represent a leap in the robot control paradigm: it transforms "hardware failures" from fatal blows into challenges that can be "circumvented."

Why Do Robots Always "Break Down"?

In the history of robotics, there is an intractable problem: the strong binding between control and the robot's body.

Service robots will completely stop working when faced with complex environments and even minor damage.

The root of the problem is that most algorithms are fine - tuned for a single piece of hardware.

They are like students who "memorize mechanically": once the problem type changes, they are at a loss. A slight change in the environment can make the robot stop.

Skild Brain aims to break this limitation. Instead of being customized for a single machine, it creates 100,000 different robot forms in a virtual world and allows the same "brain" to adapt to them.

If its four legs are sawed off, it learns to crawl on stumps; if a joint is locked, it re - plans its gait; if a wheel gets stuck, it immediately switches to walking mode; even after being forced to wear "stilts," it can stagger and regain its balance.

What's amazing about this ability is that it is almost "zero - shot," meaning these extreme damage scenarios are not in the training set.

Skild Brain can still quickly find a viable way forward in unknown situations.

Wired used an extreme scenario to describe it in a report:

Even if all four legs of a quadruped robot are cut off by a saw, it can still move forward using its torso and stumps.

This is not a horror movie plot but a real technological demonstration.

That's why Skild Brain is regarded as a paradigm shift: it turns "malfunctions" from endpoints into new starting points.

Even if the body is severely damaged, the brain can still adapt and find another way to move forward.

How Does It Do It? The "Brain Logic" of Skild Brain

After seeing the video of the damaged robot still moving forward, many people may wonder: what enables it to keep going?

Skild didn't achieve this by magic but through a combination of extreme training and design: large - scale simulation training, generalization models, and an online self - adaptation mechanism.

They built a virtual universe with 10,000 different robot forms and let the model "walk" for 1000 simulated years to force it to learn cross - form adaptation.

The purpose is to prevent the model from rote learning and make it learn a strategy that can quickly respond in unknown structures.

Specifically, this system integrates at least three key paths:

The first is reinforcement learning.

The model keeps trying, failing, and optimizing in the simulated world, just like a child learning to walk.

Failure is common, but it helps the model accumulate knowledge of strategies.

Skild Brain can remember failures and immediately correct them in the next action. This is the qualitative change brought about by the combination of reinforcement learning and long - term memory.

The second is transfer and generalization ability.

Traditional robot algorithms are often confined to a single - machine world: a robotic dog that has learned to run has to start from scratch when given a new body.

Skild Brain aims for "learn once, use everywhere" - just like humans can change shoes, use crutches, or even wear prosthetics and keep walking without having to relearn how to walk.

The third is the online self - adaptation mechanism.

Skild Brain can sense the body's state in real - time during operation - which leg is stiff, which motor is jammed, and immediately adjust the control strategy.

This "immediate adjustment" logic is like when a person is sick or injured, the brain immediately re - plans the gait and uses another leg or external support to maintain balance.

In another experiment, researchers locked two motors of a quadruped robot with wheels and legs. Instead of being paralyzed, it maintained balance like a wobbly bicycle using the remaining two wheels.

This is the amazing part of Skild Brain - it can find a new way to survive in unprecedented extreme situations.

It is this combination of logic that gives Skild Brain its "unstoppable" ability.

What sets Skild Brain apart from traditional robots is not just its "fast reaction."

Skild said that its brain has extraordinary memory: most robot control strategies have a memory window of only a few hundred milliseconds, only capable of handling one immediate action, while Skild Brain's context window is more than a hundred times longer.

This means it can "remember" feedback and action trajectories from a longer period in the past, just like humans learning to adjust their steps from experience rather than relying solely on instantaneous reflexes.

That's why when it is forced to make a quadruped robot stand and walk, it can "understand" it as a humanoid body in a zero - shot environment and temporarily adjust its gait to adapt.

From short - term reflexes to long - term memory, from hardware dependence to cross - body generalization, the logical chain of Skild Brain is gradually completed: it can not only handle immediate accidents but also constantly turn the unfamiliar into the familiar and the unexpected into the normal in the cycle of memory, experience, and adaptation.

One Brain to Rule Them All: Applications and the Future

Skild Brain is not just a technological show in the laboratory. Its significance lies in that when the "brain" can truly exist independently, robots can show "resilience" in more scenarios.

In the industrial and commercial fields, this means that robots on the production line don't have to stop the entire line due to a small malfunction.

In the past, a loose screw would lead to production suspension for maintenance; in the future, Skild Brain can make the robot keep working with the damage and wait for the maintenance window for unified repair, greatly improving efficiency and lifespan.

In disaster rescue, its value is even more obvious.

Earthquakes, fires, and mine disasters create harsh environments for robots, and they are likely to be damaged under such conditions.

The self - adaptation ability of Skild Brain allows it to continue performing search and rescue tasks until the end, even in a "severely damaged" state.

The military field is also a potential scenario.

On the battlefield, equipment damage is almost inevitable. If the robot's "brain" can find a way to survive in a damaged body, it means more endurance and survivability.

In the consumer market, the idea of a "brain" that can be "installed in different shells" is also very appealing: today it's in a vacuum cleaner, tomorrow it can be in a robotic dog, and the day after tomorrow it can be transferred to a household assistant robot.

One - time training with multiple uses can significantly reduce costs.

What really deserves in - depth thinking is that Skild Brain may not just add a few usage scenarios but represent a new robot ecosystem: one brain to rule them all.

Robots will no longer be isolated individual machines but share the same thinking core. Their actions will increasingly resemble "one brain controlling thousands of bodies."

This is not only a miracle of efficiency but also brings new challenges: who has control over this brain? If its adaptability exceeds expectations, can humans fully control it? How will the social structure and labor market be reshaped?

Just like the flashing words "Adapting…" in the Skild video. This is not only a skill learned by robots but also a portrayal of technology itself.

The question is: when an "ever - running brain" is truly born, how should humans embrace it?

Reference materials:  

https://x.com/SkildAI/status/1970940614234771579 

https://www.skild.ai/blogs/building-the-general-purpose-robotic-brain?utm_source 

This article is from the WeChat official account "New Intelligence Yuan", edited by Qingqing. Republished by 36Kr with permission.