[Model Release] Google DeepMind releases RT-2: Robotic Transformer that transfers web knowledge to robot control

Google DeepMind introduces RT-2, a novel Vision-Language-Action (VLA) model that enhances robotic control by integrating knowledge from both web-scale datasets and robotics data. Building upon the RT-1 model, RT-2 translates generalized instructions for robotic control while retaining web-scale capabilities.