The data to be translated: Google DeepMind researchers have developed a large language model called SayTap, which can convert various human instructions into a format that a quadrupedal robotic dog can understand. This model enables the robotic dog to not only comprehend basic commands such as moving forward or backward, but also to grasp more complex situational concepts, such as catching a squirrel or running quickly. Researchers indicate that SayTap, as a controller that can be transmitted to real robot hardware, bridges the gap between natural language and the underlying controller, allowing humans to command robots simply by speaking to them as they would to another person. Although this technology is still in its early stages, it holds the potential for humans to control robots through natural language instructions in the future.