Robots worldwide, including a robot from Google, are exchanging data on object manipulation to progress towards a general-purpose robotic brain operating system. Generative AI tools, such as ChatGPT and Midjourney, rely on massive neural networks trained on a large dataset pulled from the web. However, the success of these AI capabilities has not translated to practical and broadly useful robots that are so common in science fiction. The Internet’s lack of robotic-interaction data hinders robots from centralized learning, thus impacting real-world performance. The RT-X project, uniting 32 robotics labs, aims to gather data, resources, and code to create centralized general-purpose robots.
This collaborative effort seeks to control various types of robots with a single deep neural network and to improve performance based on shared experiences. The RT-X dataset contains about a million robotic trials for 22 robot types and numerous behaviors, fostering the creation of more advanced and generally useful robots. Integrating web-scale knowledge helps robots understand complex tasks, and the inclusion of multirobot RT-X data enhances the robot’s ability to generalize. The RT-X project indicates that with a unified effort, a general robotic brain operating system capable of universal robotic control and data sharing can be created and is close to reality.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…