Atlas sorts car parts: Boston Dynamics tests robot in Hyundai factory

Contents

Boston Dynamics' robotics researchers are showcasing their latest achievement: In Hyundai's production facilities in the US state of Georgia, the Atlas robot is demonstrating for the first time under real working conditions how it performs tasks autonomously. The cameras of the US television magazine 60 Minutes accompanied the humanoid machine and showed impressive footage.

At Boston Dynamics, there is currently a high-pressure effort to win the race for the first fully automated industrial workers. Because one thing is clear: the competition – especially from China – is not sleeping.

Atlas: from a jumping robot to a factory worker

Atlas is about 1.75 meters tall and weighs about 90 kilograms; its mechanics and control system have been designed to function on the factory floor.

The humanoid robot has been part of Boston Dynamics' robotics research for years, but the current model differs fundamentally from previous versions, which primarily demonstrated spectacular movements like jumps.

In the current test, Atlas completed a task in the spare parts warehouse of the Hyundai factory largely without human assistance – by sorting roof racks for the assembly line, for example.

Atlas learns from humans and then finds its own ways

Boston Dynamics engineer Zack Jackowski explains in the television report that Atlas works autonomously and has been specifically trained to perform tasks in a real production environment. The crucial point: not every single step is pre-programmed. The robot finds its own solutions and strategies for various tasks.

To achieve this, the engineers combine intensive training phases in the lab with AI-supported learning: first, humans demonstrate movements to the robot, which serve as a basis for machine learning. This creates models that Atlas can later use to perform similar tasks independently and develop individual solutions.

Inside the robotics lab, movement data is partially recorded via teleoperation, where humans guide the robot through specific sequences using virtual reality systems. This data is fed into the AI models so that Atlas can become operational without direct human control later on.

Learning through simulation

In addition to teleoperation, the developers use modern machine learning methods. Hundreds of virtual instances of Atlas practice movements simultaneously and learn from mistakes before they are deployed in the real environment.

Furthermore, the team simulates movement data so that Atlas develops a broad basis of action perspectives before undertaking real physical tasks.

The result of this development work is a robot that not only masters physical tasks such as grasping and handling. It also learns from experience how complex physical processes can be successfully carried out in an industrial environment.

AI advances humanoid robotics

With Atlas, Boston Dynamics aims to secure its role as a pioneer in a global race. In addition to established players, companies like Tesla and various startups from the USA and China are competing in the industry.

Atlas remains a research project for now and is not yet ready for industrial deployment. According to Boston Dynamics, it could still take several years before the humanoid robot is fully deployed in industry.

This article first appeared on t3n.de.

(jle)