Boston Dynamics’ Atlas Robot Takes on Factory Tasks Without Human Control

Boston Dynamics recently shared a new video showcasing the abilities of its humanoid robot, Atlas, and this time, it’s working without any human control.

Unlike previous videos that highlighted the robot’s wide range of movements, this latest clip reveals how Atlas can independently handle sorting tasks in a simulated factory setting using its machine learning skills and enhanced sensors.

In this demo, Atlas takes on the task of moving plastic engine covers between “supplier containers and a mobile sequencing dolly,” which highlights the robot’s multiple features working seamlessly together. With only a “list of bin locations to move parts between,” Atlas leverages its sensors and machine learning capabilities to identify the bins and decide how to use its limbs and three-fingered hands to grab and relocate the covers.

At first glance, the nearly three-minute video might not seem too groundbreaking—just another robot handling a tedious job. But what stands out here is that Atlas is doing it all autonomously.

According to Boston Dynamics, there’s no one remotely guiding its actions or decisions, unlike the Optimus robots presented by Elon Musk and Tesla, which still require human direction. Atlas uses a machine learning (ML) vision model to detect and localize the environment.

Atlas completes each step, even adapting to changes in real-time. At one point, when it faces resistance due to a misaligned part, Atlas pulls the part out, adjusts its position, and tries again until it fits properly. Atlas’s ability to handle task failures is honestly impressive. It shows that the robot can operate in “real-world” work settings where it might encounter uneven surfaces, dented trucks, or oddly-shaped containers—and adapt accordingly. 

Throughout the video, we see Atlas tackle a cart of plastic trays, carefully maneuvering each tray with its “fingers” before moving backward, twisting its body to place them in vertical slots. There’s also a glimpse of the robot’s vision system, which analyzes each tray's size and shape before placement.

As Nvidia CEO Jensen Huang said earlier this year: “Everything is going to be robotic. All of the factories will be robotic. The factories will orchestrate robots and those robots will be building products that are robotic.” If this carries on, he may indeed be onto something big.