Robot from Boston achieves lifelike motion thanks to unified AI model

Published on Sep 05, 2025 at 10:47 AM (UTC+4)
by Jason Fan

Last updated on Sep 05, 2025 at 7:55 PM (UTC+4)
Edited by Emma Matthews

The famous Atlas robot from Boston Dynamics has taken a big leap forward, thanks to a new unified AI model that controls both its arms and legs at the same time.

This means the humanoid can now walk and grab objects with a single brain, instead of relying on separate systems for movement and manipulation.

Even more impressive, Atlas is beginning to show hints of ’emergent’ skills, which are things it was never specifically taught but figured out anyway.

For a machine once known mostly for parkour stunts and viral dance routines, this shift could be the beginning of something much more meaningful.

SBX CARS – View live supercar auctions powered by Supercar Blondie

The Atlas robot is learning quickly

So, what does that look like in practice?

Imagine Atlas reaching into a bin to grab an object.

Instead of wobbling awkwardly, the robot naturally adjusts its stance, just like a person would, planting its legs to keep balance.

If it drops the item, Atlas doesn’t freeze; it bends down and tries to pick it up again, even though no one ever programmed that move directly.

That kind of adaptability is rare in robotics and a sign that general-purpose intelligence may finally be creeping into the field.

At the very least, Atlas seems to be slightly ahead of Tesla’s own Optimus robot, which recently received some online criticism for how awkwardly it moves.

The unified AI model allows for more practical applications

The project is a collaboration between Boston Dynamics and the Toyota Research Institute (TRI).

By feeding Atlas a mix of visual input from cameras, data from its sensors, and a wide library of example actions, the team created what they call a ‘large behavior model.’

Think of it like training a language AI, but instead of learning words, Atlas is learning how to use its body in smarter, more flexible ways.

Experts say this is a major breakthrough.

“The coordination of legs and arms is a big deal,” noted UC Berkeley roboticist Ken Goldberg.

It might not sound flashy compared to kung fu moves, but being able to handle real-world tasks is what will eventually make humanoid robots practical.

After all, a robot that doesn’t need to sleep will likely be a major boon for manufacturing companies worldwide.

DISCOVER SBX CARS: The global premium car auction platform powered by Supercar Blondie

user

Jason Fan is an experienced content creator who graduated from Nanyang Technological University in Singapore with a degree in communications. He then relocated to Australia during a millennial mid-life crisis. A fan of luxury travel and high-performance machines, he politely thanks chatbots just in case the AI apocalypse ever arrives. Jason covers a wide variety of topics, with a special focus on technology, planes and luxury.