Google DeepMind Creates AI System to Train General-Purpose Robots

The system, created alongside 33 other research institutions, is designed to be a holistic training platform across robot models

Scarlett Evans, Assistant Editor, IoT World Today

October 10, 2023

2 Min Read
The system can be leveraged across robot types
Google DeepMind

Google DeepMind, alongside 33 other research institutions, has created a general-purpose AI system to train multiple kinds of robots.

The system is designed to reduce the time and labor spent on training robots for specific tasks, instead creating a holistic system for training robots at scale.

Dubbed Open X-Embodiment, the platform contains data from 22 different robot types to create a dataset resource for general-purpose robots to learn from. 

“In this work,” DeepMind said, “we show training a single model on data from multiple embodiments leads to significantly better performance across many robots than those trained on data from individual embodiments.”


The partners have also released RT-1-X, a robotics transformer (RT) model trained on the dataset which demonstrates skills transfer across robot designs. 

In tests, the DeepMind team said the platform showed “significantly better performance” across robots than those trained on individual datasets. They also found training the visual language action model, RT-2, on data from multiple embodiments “tripled its performance on real-world robotic skills.”

The Open X-Embodiment dataset has more than 500 skills and 150,000 tasks stored across more than 1 million episodes. The team says it is the most comprehensive robotics dataset of its kind.

Related:Google DeepMind Launches 4-Legged Robots Benchmark

“We developed these tools to collectively advance cross-embodiment research in the robotics community,” according to a DeepMind statement. “We believe these tools will transform the way robots are trained and accelerate this field of research.”

The Open X-Embodiment dataset and RT-1-X model checkpoint are now available for the broader research community. 

In the future, the team said they may investigate how to combine the system with self-improvement capabilities, allowing robots to learn as they operate.

About the Author

Scarlett Evans

Assistant Editor, IoT World Today

Scarlett Evans is the assistant editor for IoT World Today, with a particular focus on robotics and smart city technologies. Scarlett has previous experience in minerals and resources with Mine Australia, Mine Technology and Power Technology. She joined Informa in April 2022.

Sign Up for the Newsletter
The most up-to-date news and insights into the latest emerging technologies ... delivered right to your inbox!

You May Also Like