This isn’t a joke — it actually happened. A company known for video technology has built a general-purpose “brain” for robots.
Different from traditional specialized robot brains, MotuBrain not only has the ability to predict and simulate the world, but can also output action instructions. It truly achieves the idea of “knowing and doing as one.”
The model is called MotuBrain. In mid-April, it quietly topped two international benchmarks, yet no one knew where it came from. People in the embodied AI field speculated for three weeks.
Now, Shengshu Technology has stepped forward to claim it. Yes — the same company behind Vidu.
MotuBrain Tops Two Benchmarks Simultaneously
One benchmark tests whether a model can understand the physical world.
The other tests whether it can actually take action.
It’s like a person competing in a physics contest while also taking a forklift operation exam — and getting the highest score in both.
On the scoreboards:
- On WorldArena, MotuBrain ranked first in motion quality and motion smoothness
- On RoboTwin2.0, MotuBrain was the only model scoring above 95 in randomized environments
In the past few years, excelling in just one of these tests was already difficult.
Topping both at the same time? No one had done it before.
Now Shengshu Technology is saying: one MotuBrain model is enough.
Why MotuBrain Emerged from a Video Company
At first glance, it sounds strange — a video company building robot brains. But the logic actually runs deep.
The future of embodied intelligence requires a World Action Model. And that must be built on top of video models that understand the physical world.
For example, in a drifting car video, the model needs to understand:
- Why the car turns
- Why the tires smoke
- Where it will go next
Seen this way, it’s not surprising that MotuBrain comes from a video-first background.
MotuBrain Performance: Crushing Both Leaderboards
MotuBrain silently topped both WorldArena and RoboTwin2.0, sparking curiosity across the field.
After weeks of speculation, on April 29, Shengshu Technology finally confirmed it.
Looking back, the clues were already there.
In December 2025, Shengshu open-sourced Motus, a general foundational world model. Less than four months later, MotuBrain arrived — a fully upgraded commercial version with key capability breakthroughs.
MotuBrain on WorldArena: Understanding the Physical World
WorldArena asks:
- If you push an object, where will it go?
- What happens after two objects collide?
- Are motion trajectories smooth and realistic?
Its metrics include:
- Motion Quality
- Flow Score
- Motion Smoothness
As of April 21, MotuBrain ranked first in all three.
This shows MotuBrain achieves comprehensive leadership in physical understanding.
MotuBrain on RoboTwin2.0: Acting in the Real World
RoboTwin2.0 provides 50 tasks:
- Grasping, placing, pushing, pulling, rotating
Two environments:
- Clean (fixed conditions)
- Randomized (changing positions, lighting, angles)
MotuBrain scored:
- 95.8 (clean)
- 96.1 (randomized)
It is the only model above 95 in randomized settings.
Across tasks:
- 90% scored above 90
- Half reached 100
This is not just leading — it’s a clear gap.
MotuBrain Combines World Understanding and Action
One benchmark tests “understanding.”
The other tests “action.”
Traditionally, these are separate systems.
MotuBrain proves they can be unified in one model — a key breakthrough for embodied AI.
MotuBrain Real-World Demo: Robots That Think While Acting
From the official demo, MotuBrain shows strong real-world capability.
Three humanoid robots completed five tasks:
- Flower arranging
- Sofa tidying
- Serving hotpot
- Mixing drinks
- Organizing a wash area
One MotuBrain, Multiple Robot Types
MotuBrain works across different robot bodies and sensors.
The more robots it connects to, the better it performs.
Long-Horizon Tasks with MotuBrain
Tasks like flower arranging require continuous planning.
The robot:
- Picks flowers
- Inserts them
- Waters them
Smooth and uninterrupted — powered by MotuBrain.
MotuBrain Shows Understanding Before Action
In a hotpot scenario:
The robot checks if the ladle is empty before scooping again.
This shows MotuBrain can:
- Understand the current state
- Predict outcomes
- Adjust actions
Unlike traditional robots, it doesn’t blindly repeat.
Multi-Task Capability of MotuBrain
In drink mixing:
- One hand pours liquid
- The other pours milk
- Then combines them
- Adds garnish
This reflects MotuBrain’s multi-task generalization ability.
Why MotuBrain Works: World Action Model
There are three main approaches:
- Direct action (VLA)
- Predict then act
- MotuBrain’s approach: predict and act together
Advantages of MotuBrain:
- Faster response
- Shared representation reduces errors
It works like human driving — prediction and action happen at the same time.
MotuBrain Core Technology: Unified Modeling
MotuBrain builds on Motus:
- Unifies video and action
- Uses a shared representation system
It enables:
- Vision-language-action integration
- World modeling
- Video generation
- Inverse dynamics
- Joint prediction
All learned together inside MotuBrain.
Practical Strengths of MotuBrain
MotuBrain solves real-world challenges:
- Works with different camera setups
- Understands natural language
- Transfers across robots
- Handles long tasks
Its performance improves with task diversity — a key advantage.
MotuBrain and Vidu: A Dual Strategy
Shengshu’s strategy includes:
- Digital world → Vidu
- Physical world → MotuBrain
Both share the same technical foundation.
MotuBrain’s Competitive Advantage
Most robotics companies lack video data.
Most video companies lack action data.
MotuBrain benefits from both.
This combination creates a strong moat.
MotuBrain and the Future of Robot Brains
The industry focus is shifting:
- From robot bodies → to robot brains
Capital is flowing into companies building systems like MotuBrain.
They are competing for the future interface to the physical world.
At this moment, MotuBrain stands out with dual benchmark dominance.
While others debate approaches, MotuBrain shows a unified path is possible.
If video models help AI understand the world,
then MotuBrain represents the step into actually acting in it.
