Robot folds laundry, packs parcels - and improvises in the process

Kevin Hofer

7-4-2026

Translation: machine translated

The robotics company Generalist has developed GEN-1, a new AI model for physical tasks. It independently compensates for disruptions and masters movement sequences that it has not explicitly learnt.

GEN-1 achieves according to Generalist a success rate of 99 per cent for precise, recurring tasks - such as folding boxes, packing smartphones or servicing robotic vacuum cleaners. It works three times faster than its predecessor GEN-0, and GEN-1 is ready for use after one hour of adaptation to robot-specific movement data.

Many hours of training data

GEN-1 is based on GEN-0, which Generalist presented in November as a feasibility study for scaling laws in robotics training. The main problem: unlike language models, which are trained on trillions of words from the internet, robotics models lack a comparable data source.

Generalist solves this with «Data Hands» - wearable grippers that record micro-movements and visual information while humans perform manual tasks. The company collected over half a million hours and petabytes of interaction data.

The result: a system that accurately inserts banknotes into a wallet, folds laundry or sorts car parts.

Improvisation instead of rigid programming

GEN-1 differs from previous robotics solutions in its ability to improvise and learn from previous experience - even in the event of faults that lie far outside the training data.

A video shows how the robotic hands can refold a displaced shirt or grasp and insert car parts after they have slipped out of place.

«Nobody has programmed the robot to learn from mistakes - and yet it does», says generalist researcher Felix Yanwei Wang in another video. «He also does it for free.»

No lone wolf in the robotics race

Generalist is not alone in the race for machine learning in the physical world. In 2025, Google presented the «Visual Learning Action» capabilities of its Gemini Robotics models that understand and implement general instructions. Physical Intelligence caused a stir with robotic arms on a driving platform that were trained in simulated households and can perform tasks such as mopping or making beds.

Tesla is also working on the technology with the humanoid Optimus robot. However, the company is not yet as advanced as Generalist or Google - the robots will still be controlled by humans in 2024.

Generalist sees GEN-1 as a turning point, comparable to GPT-3. Some tasks already exceed the threshold for commercially viable use - and more complex tasks are set to follow with each new model generation.

Kevin Hofer

Senior Editor

kevin.hofer@digitecgalaxus.ch

From big data to big brother, Cyborgs to Sci-Fi. All aspects of technology and society fascinate me.

---

News + Trends

From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.

Show all

These articles might also interest you