“The PaLM-E and Robotic Transformer 2 (RT2) projects showed that a powerful vision-language model can be adapted with robotics data to generate low-level robot actions.”