.jpg?w=1024)
Gripping a power drill was the easy part. Aligning it with a bolt proved harder.
WIRobotics said its upper-body ALLEX humanoid achieved a 100% success rate in drill-grasping and actuation tests, but only 50% when aligning the drill with a bolt. The demonstration, conducted through the Physical AI Fellowship backed by AWS, NVIDIA, and MassRobotics, highlights both the progress and remaining limitations of AI-powered robots.
The results suggest robot foundation models are improving at specific physical tasks, although precision and repeatability remain barriers to broader deployment.
The practical ceiling for robots
The sim-to-real gap has been one of the biggest challenges for companies in the robotics sector. This gap, which defines the level of accuracy between simulation and real-world conditions, can be narrowed by using more data from the physical world.
The ALLEX (ALL-EXperiences) robot uses NVIDIA's GR00T N1.7 robotic foundation model, which was trained on a combination of real captured human data, real robot demonstrations, and synthetic data. This combination of training data made it possible to for ALLEX to achieve a high degree of accuracy on physical tasks.
What WIRobotics achieved
ALLEX’s ability to properly use a drilling machine is more than a reflection of its capability to handle household tools.
The humanoid robot was able to accomplish its goals with human-level dexterity and full body control across the hands, arms, neck and waist. This achievement signaled that ALLEX could in fact operate in a real environment.
Though, it's important to note that ALLEX is currently an upper body humanoid robot. There's no lower body.
From the announcement blogpost, the company's team has shared that this drilling skill is only one of the many capabilities that they intend to add to a growing list for “tasks such as assembly, maintenance, inspection, and machine operation.”
How enterprises can prepare for the next chapter
Robots currently deployed in industries like manufacturing can only perform pre-determined repetitive tasks. However, when met with an unstructured problem, the robots fail.
This is where intelligent robotic agents like ALLEX come in, being able to navigate real-world human challenges in a dynamic environment.
The car-making giant Hyundai has already laid out plans to use Boston Dynamics’ Atlas humanoid robots in their factories by 2028.
This action is in sharp contrast to how humanoid robots have mainly been “just for show” in recreational settings like dance events.
Hyundai further highlighted that the “whole-body coordination and object manipulation, including the delivery of the ceremonial match ball during a FIFA World Cup 2026™ match and the lifting and transport of a 23-kilogram compact refrigerator” were validation that the robots were ready for industrial use.
The question for enterprises is no longer whether robots will be in the production workforce, but when. Even though robots still need more fine tuning before they enter the workforce, enterprises can keep abreast with the free resources provided by industry leaders like NVIDIA and Google.
Read more: Learn how humanoid robots are navigating real-world workplaces and tackling practical tasks beyond controlled environments.