Watch this humanoid robot nail complex tasks and think on the fly

By Trevor Mogg Published August 20, 2025

Boston Dynamics’ humanoid robot may have skipped the inaugural “robot Olympics” in China last week, but that doesn’t mean the engineers behind the machine have been sitting around watching the world go by. 

Indeed, a video released by the Massachussetts-based company on Wednesday reveals that the team has been hard at work on Atlas, its advanced and highly talented bipedal bot.

Working with experts at the AI- and robotics-focused Toyota Research Institute (TRI), Boston Dynamics has equipped Atlas with a Large Behavior Model (LBM), essentially a sophisticated AI system trained on vast datasets of human actions, aimed at enabling the robots to understand, generate, and adapt complex, human-like behaviors for activities in real-world environments.

The video shows Atlas performing a lengthy sequence of complex tasks that force it to combine object manipulation with locomotion. They include walking, crouching, and lifting objects, while at the same time packing, sorting, and organizing. 

“By adopting LBMs, new capabilities that previously would have been laboriously hand-programmed can now be added quickly and without writing a single new line of code,” the Massachussetts-based company said in a release.

To test its ability to adjust itself, an engineer interrupts Atlas in the way that an annoying co-worker might do, by repeatedly closing the lid of the box from which it’s taking things, and by sliding the box across the floor. 

If Atlas had a voice — and no doubt one day it will —  it would probably have said: “Can you quit messing around — I’m trying to get a job done here.”

Atlas passes with flying colors, refraining from decking the troublemaker and instead readjusting its position to continue with the task in hand. 

“This work provides a glimpse into how we’re thinking about building general-purpose robots that will transform how we live and work,” said Scott Kuindersma, Boston Dynamics’ vice president of robotics research. “Training a single neural network to perform many long-horizon manipulation tasks will lead to better generalization, and highly capable robots like Atlas present the fewest barriers to data collection for tasks requiring whole-body precision, dexterity, and strength.”

Boston Dynamics is one of a growing number of tech firms working on humanoid robots, with rapidly advancing technology paving the way for increasingly agile, dexterous, and intelligent bipedal robots that could one day perform a huge variety of activities. Even the laundry …