Alphabet Inc
Robotics Transformer 2, or RT-2, is a first-of-its-kind vision-language action (VLA) model that combines text and image data from the web and can directly produce robotic actions.
Like language models that learn general concepts from web text, RT-2 transfers this knowledge to guide robot behavior, making it a pioneering development in robotics and AI.
Until now, robots operated through complex systems where high-level reasoning and low-level manipulation were disconnected, resulting in a cumbersome process similar to relaying messages throughout the body. However, RT-2 changes this approach by integrating sense and action within a single model.
This breakthrough allows RT-2 not only to perform intricate reasoning tasks but also to generate robot actions directly. The most notable aspect is that it can transfer knowledge from its language and vision training data to handle new tasks, even without specific training.
For instance, previous systems required explicit training to identify and dispose trash. In contrast, RT-2 leverages its vast corpus of web data to understand and remember what trash is effortlessly.
Furthermore, it can intuitively know how to dispose of the trash, despite never receiving explicit training for that specific action.
Even when dealing with abstract concepts, like trash changing form from a bag of chips to a banana peel after consumption, RT-2 can comprehend the context from its vision-language training and successfully perform the task.
While Google does not look to commercialize the latest AI model, they could gradually find a place in warehouses or as home assistants, the New York Times reports.
In June, Uber Technologies, Inc
Price Action: GOOG shares traded higher by 3.00% at $133.77 on the last check Friday.
