Skip to main content
Media data LogisticsInnovation
LOGISTICS NEWS SWITZERLAND

The news platform for Switzerland, the EU and the rest of the world




Understanding instead of just "seeing"

March 16, 2026

Chaotic storage is still considered a prime example for those too lazy to search. At the Technical University of Munich (TUM), a robot has now been developed whose primary task is to track down missing items. According to TUM, it's one of the first robotic companions that not only integrates image understanding but also meaningfully connects it. 

TUM Search Robot 1 310

a tracker with RFID, Wi-Fi, and a barcode to every conceivable object and item is clearly not the obvious solution here. Instead, the robot from Professor Angela Schoellig's TUM Learning Systems and Robotics Lab, which has a camera mounted on its top,  combines information from the internet with a spatial map of its environment.

To find something, the robot has to look around and create a three-dimensional image of the room. The camera initially delivers two-dimensional images, but their pixels also contain depth information.

Pixels with depth information

This creates a three-dimensional image of the environment accurate to the centimeter, which is constantly updated. A laptop also provides the robot with information about which objects are visible in the image and what significance they have for humans.

"We taught the robot to understand its surroundings."

Angela Schoellig, head of the robotics lab at the TUM Chair of Safety, Performance and Reliability for Learning Systems, envisions developing robots that can navigate autonomously in any environment. Humanoid robots working in factories, or robots in care settings that inhabit different homes, require this newly developed fundamental understanding, which, as Schoellig says, is "important for all robots that move in constantly changing spaces"—that is, in dynamic environments

TUM Search Robot 2 310Photos: TUM

Based on internet information, the robot understands that a table or windowsill could be used to briefly set down a specific item, while a stovetop or sink would be unsuitable. "The language model incorporates the relationships between the objects, and we translate this information into the robot's language"explains Schoellig. Small two-digit numbers appear on the three-dimensional map of the environment, constantly recalculating the probability of the sought-after object being located there.

In a dynamic environment

The robot then searches the likely locations almost 30 percent more efficiently than if it searched randomly around the room, according to the research findings. Artificial intelligence is thus used in two ways: firstly in image recognition, and secondly through the use of a language model. A demonstration of the search robot was on display at the German Robotics Congress in Cologne until Friday.

https://robotics-insti...event/grc/

LI-NEWSLETTER

Every week, the latest logistics news from Switzerland, the EU and the rest of the world.

Subscribe now

You might also be interested in this

Protect autonomous vehicles

October 9, 2026

Combining road and rail in a modular way

August 6, 2026

Securing a technological advantage

March 17, 2026

LI-Shop


WAGNER Switzerland AG




Impressive study trip

ils study trip video cover image

Around 50 participants took part in the ILS study trip in 2025, which provided interesting insights into state-of-the-art logistics facilities.

Read more at...

Video camera



With 30,000 visitors

Monika Erb on Video 310

With almost 30,000 visitors and a chorus of honking horns, the 13th edition of transport.ch at Bernexpo came to a close. For four days, it was the focal point of the industry.

Read more at...

Video camera



Partner forklift trucks

Partner Intralogistics

Partner Associations

Partner Trade Fairs & Events